feat(docs): add management summary link in evaluation documents

- include link to one-page executive summary in README
- update dashboard section in evaluation HTML to reference summary
- revise report metadata to reflect additional CSV tables generated

.cursorrules

@@ -0,0 +1,59 @@
+# Cursor rules for LabDataStorageEvaluation
+
+## Commit messages (mandatory format)
+
+Every commit message you generate MUST follow this convention exactly.
+
+### Source of truth (staged diff only)
+
+Base the message ONLY on `git diff --cached` (the index / staged changes).
+Do not use unstaged working-tree diffs, untracked files, recent commit
+messages, or chat context.
+
+- Mention only paths that appear in the staged diff.
+- If the index is empty, output exactly: `No staged changes`
+- Never mention a file (e.g. README.md) that is not in the staged diff.
+
+First line: type(scope): description
+- lowercase imperative English summary, under 72 characters, no trailing period
+- type is one of (closed list): feat fix perf refactor security docs ci chore test
+- scope is one of (closed list): specs rules docs config datagen convert bench report tools all
+
+Scope selection by changed files:
+- specs   = docs/specs/**
+- rules   = docs/rules/**
+- docs    = other prose docs (README.md, CLAUDE.md, docs/examples/**)
+- config  = configuration files (.gitignore, .cursorrules, editable YAML defaults)
+- datagen = lab-configuration and data-generation code (tasks 01-03:
+  make_lab_config.py, make_process_flow.py, generate_data.py)
+- convert = format converters: JSON-LD, SQLite, PostgreSQL, RDF (tasks 04-07)
+- bench   = benchmark queries, runner, extrapolation (tasks 08-10)
+- report  = reporting code and templates (task 11)
+- tools   = helper scripts under tools/
+- all     = multiple areas, or when no narrower scope honestly fits
+
+Body: the subject is only a title. If the diff touches more than one area,
+add a blank line and 2-6 concise bullets covering every meaningful changed
+area. Split unrelated work into separate commits.
+
+Special forms allowed as-is: "Merge ..." commits; "release(scope): vX.Y.Z".
+
+Never emit vague or unprefixed subjects such as: "Update files",
+"Improve project", "clarify rules", "changes".
+
+Examples of good subjects:
+- feat(datagen): stream friction cycle rows during CSV generation
+- fix(convert): match sqlite row counts against MANIFEST before commit
+- docs(rules): add benchmark methodology rule
+- chore(config): ignore out/ artifact tree
+
+## Typography
+
+Use plain ASCII typography in everything you write into this repository:
+hyphen-minus instead of en/em dashes, straight quotes, three dots instead
+of the ellipsis character. See docs/rules/code-data-formatting.md.
+
+## Python
+
+Python code in this repository uses TABS for indentation. See
+docs/rules/code-python-style.md for the closed dependency list.

.gitignore

@@ -0,0 +1,21 @@
+# Generated pipeline artifacts - reproducible from specs + rules + code
+# (see docs/rules/build-pipeline-tasks.md)
+out/*
+!out/report/
+!out/report/**
+!out/config/
+!out/config/**
+!out/.done/
+!out/.done/**
+
+# Python
+__pycache__/
+*.pyc
+.venv/
+venv/
+
+# Editor / OS noise
+.vscode/
+.idea/
+Thumbs.db
+.DS_Store

CLAUDE.md

@@ -0,0 +1,112 @@
+# LabDataStorageEvaluation - Project Rules
+
+A reproducible evaluation pipeline for laboratory data storage formats,
+built around a simulated tribology laboratory (Sandia Pt-Au LDRD context).
+The pipeline (tasks 01-11, specs in [docs/specs/](docs/specs/)):
+
+1. Defines the lab model (`lab_config.yaml`) and process-flow diagrams.
+2. Generates **~150 MB of physically plausible measurement data in CSV**
+   (the canonical raw format), deterministically from **seed 20260711**.
+3. Converts the corpus into 4 target formats: **JSON-LD** (full + hybrid),
+   **SQLite**, **PostgreSQL 16** (indexed + partitioned), **RDF
+   triplestore** (oxigraph).
+4. Benchmarks **7 retrieval scenarios (Q1-Q7) x 5 formats**, measuring
+   wall time, peak RAM, CPU, disk I/O, and footprint.
+5. Extrapolates to 600 GB / 1.2 TB / 6 TB and produces a ranked report
+   (`REPORT.md` + charts).
+
+Everything runs **locally** (Windows development host, local PostgreSQL 16,
+no cloud services, no server deploy). The stack is **Python 3.11+ (tabs for
+indentation)** with a closed dependency list; there is **no UI, no HTTP
+API, no authentication** in this project. All artifacts land under
+`./out/` (git-ignored); tasks signal completion via `./out/.done/<NN>.ok`
+markers.
+
+These rules are **binding** for every contributor (human or AI agent)
+touching this repository. They override any pattern found in imported
+specs, external tutorials, or older project docs that predate this file.
+
+---
+
+## Mandatory rules - load before starting matching work
+
+When the user's request matches a topic below, **before writing any code or
+producing a plan**, run `Read` on the listed rule file(s). The rules are
+binding; ignoring them produces broken work that the user will reject.
+
+| If the task involves... | You MUST first Read |
+|---|---|
+| Writing or changing any Python (style, tabs, dependency list, streaming discipline) | [docs/rules/code-python-style.md](docs/rules/code-python-style.md) |
+| Data generation, physical models, random values, seeds, reproducibility | [docs/rules/data-determinism.md](docs/rules/data-determinism.md) |
+| Naming entities/IDs, timestamps, measurement columns, units, `au_wtpct_mean` | [docs/rules/data-naming-units.md](docs/rules/data-naming-units.md) |
+| SQLite / PostgreSQL schema, DDL, keys, indexes, bulk loading, `track_summary` | [docs/rules/db-sql-schema.md](docs/rules/db-sql-schema.md) |
+| Benchmark queries, the runner, resource measurement, timeouts, extrapolation, scoring | [docs/rules/bench-methodology.md](docs/rules/bench-methodology.md) |
+| The `./out/` tree, `.done` markers, task order/parallelism, `storage_sizes.csv`, re-runs | [docs/rules/build-pipeline-tasks.md](docs/rules/build-pipeline-tasks.md) |
+| Validation: MANIFEST, row counts, result checksums, expected volumes, pytest | [docs/rules/test-pipeline-validation.md](docs/rules/test-pipeline-validation.md) |
+| Python error handling (`try` / `except` / `raise`, abort semantics) | [docs/rules/code-error-handling.md](docs/rules/code-error-handling.md) |
+| Logging, `print()` vs logger, log levels | [docs/rules/obs-python-logging.md](docs/rules/obs-python-logging.md) |
+| Configuration values: the YAML configs, constants, the PostgreSQL DSN, CLI flags | [docs/rules/code-config-yaml.md](docs/rules/code-config-yaml.md) |
+| Writing, editing, or reviewing comments in source code (any language) | [docs/rules/code-comments-policy.md](docs/rules/code-comments-policy.md) |
+| Prose strings in source/config/docs (typography / encoding / dashes) | [docs/rules/code-data-formatting.md](docs/rules/code-data-formatting.md) |
+| Adding or renaming a rule file in `docs/rules/` | [docs/rules/meta-rules-naming.md](docs/rules/meta-rules-naming.md) |
+| Anything where misinterpretation could be destructive (deleting/regenerating `./out/`, dropping the `tribo` DB, changing the seed or `lab_config.yaml`, editing specs, git history) | [docs/rules/meta-no-assumptions.md](docs/rules/meta-no-assumptions.md) |
+| Every response (length / formatting discipline) | [docs/rules/meta-concise-answers.md](docs/rules/meta-concise-answers.md) |
+| Asking the user any question | [docs/rules/meta-concise-questions.md](docs/rules/meta-concise-questions.md) |
+| Every response (communication language / register - Russian, no jargon or slang) | [docs/rules/meta-language-style.md](docs/rules/meta-language-style.md) |
+| Any git operation that would create a commit, tag, merge, revert, rebase, cherry-pick, or push | [docs/rules/meta-git-commits.md](docs/rules/meta-git-commits.md) |
+| Generating commit messages via Cursor, or maintaining the `.cursorrules` artifact | [docs/rules/meta-cursor-rules.md](docs/rules/meta-cursor-rules.md) |
+
+Multiple matches -> Read all matching rule files. Unsure -> Read the closest one;
+over-loading is cheap, missing a rule is expensive.
+
+---
+
+## Other binding constraints
+
+These are short, cross-cutting constraints. They are still **binding**; the
+linked rule file holds the detailed spec.
+
+- **Determinism is the foundation.** Seed `20260711` lives in
+  `lab_config.yaml` and is never hardcoded; only seeded RNGs; no
+  wall-clock values inside the corpus; regeneration is byte-identical
+  (proof: `MANIFEST.csv` sha256). See
+  [docs/rules/data-determinism.md](docs/rules/data-determinism.md).
+- **A wrong artifact is worse than no artifact.** Validation gates (row
+  counts vs MANIFEST, result checksums with 1e-9 tolerance, the
+  120-200 MB size gate) abort loudly; a `.done` marker is written only
+  after full success. See
+  [docs/rules/code-error-handling.md](docs/rules/code-error-handling.md)
+  and [docs/rules/test-pipeline-validation.md](docs/rules/test-pipeline-validation.md).
+- **Tasks communicate only through `./out/` artifacts.** The tree is
+  git-ignored; sources of truth are specs + rules + code. Benchmarks run
+  on a quiet machine. See
+  [docs/rules/build-pipeline-tasks.md](docs/rules/build-pipeline-tasks.md).
+- **Measurement honesty.** Subprocess isolation per cell, 1 warm-up + 3
+  measured runs, randomized order, 30-min timeout recorded as `TIMEOUT`,
+  failed checksums invalidate the measurement, projected numbers are
+  always labeled. See
+  [docs/rules/bench-methodology.md](docs/rules/bench-methodology.md).
+- **Python: tabs, closed dependency list.** Stdlib first; only the
+  packages listed in
+  [docs/rules/code-python-style.md](docs/rules/code-python-style.md);
+  task 03 is stdlib + numpy only; stream everything bulk.
+- **Commit message format (mandatory).** First line
+  `type(scope): description`, lowercase imperative English, under 72
+  chars. Types (closed): `feat fix perf refactor security docs ci chore
+  test`. Scopes (closed): `specs rules docs config datagen convert bench
+  report tools all`. The AI never creates commits - git is read-only for
+  agents; the user runs all git write commands. See
+  [docs/rules/meta-git-commits.md](docs/rules/meta-git-commits.md).
+
+---
+
+## Canonical source of truth
+
+If this file and any `docs/rules/<name>.md` disagree, **the rule file wins**
+within its topic - that is where the detailed binding spec lives. If a rule
+and a task spec in `docs/specs/` disagree on a convention, **the rule wins**
+and the spec is updated to match; the specs remain authoritative for task
+scope, inputs, outputs, and dependencies. If this file and any other
+document (README, older docs, code comments) disagree, **this file plus the
+rule files win**. Update the other document to match; do not work around
+the rules.

README.md

@@ -1,11 +1,342 @@
 # LabDataStorageEvaluation
 
-Build a complete, reproducible pipeline that:
+A reproducible pipeline that answers one practical question: **which storage
+format should a materials-science laboratory use for its measurement data?**
 
+The pipeline simulates a tribology laboratory (Pt-Au LDRD context, Sandia
+National Laboratories),
+generates ~150 MB of physically plausible raw measurement data in CSV,
+converts the corpus into four additional storage formats, benchmarks seven
+laboratory-standard retrieval scenarios against all five formats, extrapolates
+the measurements to 600 GB - 6 TB, and produces a ranked decision report for
+five laboratory use cases: analysis, reporting, search/filtering, archiving,
+and inter-lab exchange.
 
-Defines a simulated tribology laboratory (Sandia Pt-Au LDRD context) with real instruments and a realistic test workflow.
+**Results - interactive, single-file HTML (open in a browser or download and share):**
-Generates ~150 MB of physically plausible measurement data in CSV (the canonical raw format).
+
-Converts the CSV corpus into 4 target storage formats: JSON-LD, SQLite (optimized), PostgreSQL (indexed + partitioned), RDF TripleStore.
+- **[Storage Decision](https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision.html)** -
-Benchmarks 7 laboratory-standard retrieval scenarios (Q1–Q7) against all 5 formats, measuring wall time, peak RAM, CPU utilization, disk I/O, and disk footprint.
+  one-page executive summary: the problem, the verdict (relational vs graph
-Extrapolates measured results to 600 GB and beyond (0.6 / 1.2 / 6 TB) using complexity-aware scaling models.
+  database), and what it costs - written for management, every claim linked
-Produces summary tables and charts that clearly rank storage formats for laboratory use (analysis, reporting, search/filter, archiving, inter-lab exchange).
+  to the full study
+  (source: [docs/research/Tribology_Storage_Decision.html](docs/research/Tribology_Storage_Decision.html)).
+- **[Storage Format Evaluation](https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html)** -
+  the full study results: dashboard, measured matrix, projections to 6 TB,
+  hardware bill of materials, weighted scoring and verdict
+  (source: [docs/research/Tribology_Storage_Evaluation.html](docs/research/Tribology_Storage_Evaluation.html)).
+- **[Foreign-Key Architecture Study](https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_FK_Architecture.html)** -
+  the measured key-schema study behind the relational design decision
+  (source: [docs/research/Tribology_FK_Architecture.html](docs/research/Tribology_FK_Architecture.html)).
+
+Reference documents:
+
+- Task specifications: [docs/specs/](docs/specs/) (plan + tasks 01-11).
+- Binding conventions: [docs/rules/](docs/rules/) and [CLAUDE.md](CLAUDE.md).
+- Full decision report (markdown): [out/report/REPORT.md](out/report/REPORT.md).
+
+## Goals
+
+1. Define a simulated tribology laboratory with real instruments and a
+   realistic test workflow.
+2. Generate ~150 MB of physically plausible measurement data in CSV (the
+   canonical raw format), deterministically from seed `20260711`.
+3. Convert the CSV corpus into 4 target storage formats: JSON-LD, SQLite
+   (optimized), PostgreSQL 16 (indexed + partitioned), RDF triplestore.
+4. Benchmark 7 retrieval scenarios (Q1-Q7) against all 5 formats, measuring
+   wall time, peak RAM, CPU utilization, disk I/O, and disk footprint.
+5. Extrapolate measured results to 600 GB / 1.2 TB / 6 TB with
+   complexity-aware scaling models.
+6. Produce summary tables, charts, and a weighted scoreboard that rank the
+   formats per use case.
+
+## Pipeline
+
+```mermaid
+flowchart TD
+    T01[01_lab_configuration] --> T02[02_process_flow_diagrams]
+    T01 --> T03[03_csv_data_generation]
+    T03 --> T04[04_convert_json]
+    T03 --> T05[05_convert_sqlite]
+    T03 --> T06[06_convert_postgresql]
+    T03 --> T07[07_convert_rdf]
+    T04 --> T08[08_benchmark_queries]
+    T05 --> T08
+    T06 --> T08
+    T07 --> T08
+    T03 --> T08
+    T08 --> T09[09_benchmark_execution]
+    T09 --> T10[10_extrapolation_model]
+    T10 --> T11[11_reporting]
+    T09 --> T11
+```
+
+Execution order: `01 -> 02 -> 03 -> (04 | 05 | 06 | 07) -> 08 -> 09 -> 10 -> 11`.
+Tasks 04-07 are independent and may run in parallel; task 09 requires a quiet
+machine (no parallel work). Every task writes a completion marker
+`./out/.done/<NN>.ok` and validates its dependencies' markers before starting.
+
+| Task | Script | Main output | Status |
+|---|---|---|---|
+| 01 Lab configuration | `make_lab_config.py` | `out/config/lab_config.yaml` | implemented |
+| 02 Process flow diagrams | `make_process_flow.py` | `out/report/process_flow.md`, `out/report/diagrams/` | implemented |
+| 03 CSV data generation | `generate_data.py` | `out/csv/` tree + `MANIFEST.csv` | implemented |
+| 04 Convert to JSON-LD | `convert_json.py` | `out/json/full/`, `out/json/hybrid/dataset.jsonld` | implemented |
+| 05 Convert to SQLite | `convert_sqlite.py` | `out/sqlite/tribo.db` | implemented |
+| 06 Convert to PostgreSQL | `convert_postgresql.py` | live PostgreSQL DB (+ dump when pg_dump is available) | implemented |
+| 07 Convert to RDF | `convert_rdf.py` | `out/rdf/dataset.nt.gz`, `dataset.ttl`, oxigraph store | implemented |
+| 08 Benchmark queries | `make_queries.py` | `out/bench/queries/`, `out/bench/expected/` | implemented |
+| 09 Benchmark execution | `bench_runner.py` | `out/bench/results_raw.csv`, `results_median.csv` | implemented |
+| 10 Extrapolation | `extrapolate.py` | `out/bench/extrapolation.csv`, `hardware_sizing.csv` | implemented |
+| 11 Reporting | `report.py` | `out/report/REPORT.md`, charts, tables | implemented |
+
+## The simulated laboratory
+
+Ti-6Al-4V coupons (10 x 10 x 3 mm) with a sputtered Cr adhesion layer and a
+Pt-Au coating deposited as a composition gradient (film thickness
+0.3-1.1 um). Four deposition batches (B721-B724) vary gun tilt, power, and
+discharge voltage. Instruments: RAPID 6-probe tribometer (friction), Bruker
+TI980 (nanoindentation), Bruker M4 Tornado micro-XRF (composition), Kurt J.
+Lesker PVD 200 (deposition), AFM (roughness), optical profilometry
+(thickness), SIMTRA (sputter transport simulation).
+
+```mermaid
+flowchart LR
+    B[Batch x4] --> W[Wafer x3 per batch]
+    W --> C[Coupon x49 per wafer, 7x7 grid]
+    C --> X[xrf_map: 400 points]
+    C --> N[nanoindentation: 25 indents]
+    C --> A[afm: Ra / Rq]
+    C --> P[profilometry: 100 points]
+    C --> T[Track x3, friction coupons only]
+    T --> CY[cof_vs_cycle: 1000 cycles]
+    T --> L[friction_loops: 10 x 200 points]
+    T --> WR[wear: 1 row]
+```
+
+| Entity | Count |
+|---|---|
+| Batches | 4 |
+| Wafers | 12 |
+| Coupons | 588 (480 friction-tested + 108 reserve/QA) |
+| Tracks | 1,440 |
+| Friction cycle rows | 1,440,000 |
+| Friction loop points | 2,880,000 |
+| XRF map points | 235,200 |
+
+Friction runs R1/R2 execute in Lab Air, R3/R4 in Dry N2. Generated values
+follow explicit physical models (COF run-in exponential, steady-state COF and
+hardness vs Au content, Archard wear), so benchmark queries return physically
+meaningful results.
+
+## Storage formats under evaluation
+
+| # | Format | Role in the comparison |
+|---|---|---|
+| 1 | CSV (raw tree) | Canonical instrument output; baseline |
+| 2 | JSON-LD (full + hybrid) | Ontology-annotated exchange format (FAIR) |
+| 3 | SQLite | Single-file relational, optimized for size and speed |
+| 4 | PostgreSQL 16 | Production relational: indexed, partitioned |
+| 5 | RDF triplestore (oxigraph) | Semantic-web store queried via SPARQL |
+
+## Benchmark scenarios (Q1-Q7)
+
+| ID | Scenario | Access pattern |
+|---|---|---|
+| Q1 | COF vs cycle curve for one track | Point read |
+| Q2 | Steady-state COF for coupons with mean Au = 10 +/- 0.5 wt% | Indexed filter |
+| Q3 | Hardness vs steady-state COF across all batches | Indexed join |
+| Q4 | Tracks in Dry N2, 100 mN, cof_ss > 0.20 (anomaly filter) | Indexed filter |
+| Q5 | Mean run-in cycles per batch over ALL raw cycle rows | Forced full scan |
+| Q6 | Wear volume vs load per coupon | Indexed join |
+| Q7 | Stribeck-style mean COF by (environment, speed-load bucket) | Full aggregation |
+
+Protocol: every cell (format x query) runs in its own subprocess; 1 warm-up +
+3 measured runs; randomized cell order; 30-minute hard timeout recorded as
+`TIMEOUT`; every result validated against a canonical checksum before the
+measurement counts. See
+[docs/rules/bench-methodology.md](docs/rules/bench-methodology.md).
+
+## Repository layout
+
+```
+docs/
+  Initial_Prompt.md   the generative prompt behind specs 00-11 (provenance, non-binding)
+  specs/        task specifications 00-11 (WHAT to build)
+  rules/        binding conventions (HOW work is done)
+  research/     measured design studies and interactive result pages
+out/            ALL generated artifacts (reproducible; bulk git-ignored,
+                report/ config/ .done/ committed - see .gitignore):
+  config/ csv/ json/ sqlite/ pg/ rdf/ bench/ report/ .done/
+common/               shared helpers (storage_sizes.csv contract, ...)
+  queries/            Q1-Q7 implementations for csv / json / rdf formats
+make_lab_config.py    task 01 entry script (one script per task, repo root)
+make_process_flow.py  task 02 entry script
+generate_data.py      task 03 entry script
+convert_json.py       task 04 entry script
+convert_sqlite.py     task 05 entry script
+convert_postgresql.py task 06 entry script (TRIBO_PG_DSN, TRIBO_PROM_URL)
+convert_rdf.py        task 07 entry script
+make_queries.py       task 08 entry script (canonical results need TRIBO_PG_DSN)
+bench_runner.py       task 09 entry script (TRIBO_PG_DSN, TRIBO_PROM_URL)
+extrapolate.py        task 10 entry script (seeds editable out/config/hw_prices.yaml)
+report.py             task 11 entry script (REPORT.md, charts, tables, scoring)
+requirements.txt      closed dependency list (docs/rules/code-python-style.md)
+```
+
+The bulk of the `out/` tree - the ~150 MB CSV corpus, the JSON-LD variants,
+databases, stores, archives, and raw benchmark data - is **excluded from
+git**; every artifact regenerates deterministically from the specs, the
+rules, and the code. A curated results subset IS committed (decision of
+2026-07-11): `out/report/` (REPORT.md, charts, tables, diagrams),
+`out/config/` (lab_config.yaml, hw_prices.yaml), and the `out/.done/`
+markers - see `.gitignore`. [docs/Initial_Prompt.md](docs/Initial_Prompt.md)
+is the original prompt the 12 specification files were generated from - kept
+for provenance only, never authoritative: where it disagrees with the specs,
+rules, or code (e.g. the SQLite key schema, the JSON-LD vocabulary), those win.
+
+## Getting started
+
+Prerequisites: Python 3.11+ (3.12 tested). PostgreSQL 16 is required only for
+tasks 06/08/09; its DSN comes from the `TRIBO_PG_DSN` environment variable
+(default `postgresql://postgres@localhost:5432/tribo`).
+
+Windows (PowerShell):
+
+```powershell
+py -3 -m venv .venv
+.venv\Scripts\python -m pip install -r requirements.txt
+.venv\Scripts\python make_lab_config.py
+```
+
+Linux:
+
+```bash
+python3 -m venv .venv
+.venv/bin/python -m pip install -r requirements.txt
+.venv/bin/python make_lab_config.py
+```
+
+Each task script supports `--log-level {DEBUG,INFO,WARNING,ERROR}` and prints
+its protocol output (paths, counts, sizes) to stdout. A task refuses to run if
+a dependency's `out/.done/<NN>.ok` marker is missing.
+
+Note on cold-cache runs: the page-cache drop used by the cold-cache benchmark
+pass is Linux-only; on Windows hosts that pass is skipped and marked as such.
+Benchmark execution is planned for a dedicated Linux host.
+
+## Results
+
+Results of the full pipeline run of 2026-07-11 (seed 20260711, 141.8 MiB
+corpus, all 35 benchmark cells checksum-valid). The generated sources of
+truth are committed under [out/report/](out/report/) - the full decision
+document is [out/report/REPORT.md](out/report/REPORT.md); the tables below
+are copied from it. An interactive single-file companion (sidebar
+navigation, per-query and per-scale views, cost explorer) is
+[docs/research/Tribology_Storage_Evaluation.html](docs/research/Tribology_Storage_Evaluation.html).
+Bulk measurement files (`out/bench/`) are git-ignored and regenerable.
+
+### Process flow diagrams (task 02)
+
+`out/report/process_flow.md` with `out/report/diagrams/D1..D6.mermaid`:
+D1 coupon assembly, D2 characterization and batch assembly, D3 testing tree,
+D4 tribometer session sequence, D5 execution loop, D6 data hierarchy.
+
+### Measured results (task 09)
+
+Storage footprint per format (working + compressed archival):
+
+| Format | Measured | vs CSV | Archival (compressed) |
+|---|---|---|---|
+| CSV | 149 MB | 1.00x | 52.8 MB (tar.gz) |
+| JSON-LD (full) | 311 MB | 2.09x | 57.2 MB (tar.gz) |
+| SQLite | 123 MB | 0.83x | 55.1 MB (gzip) |
+| PostgreSQL | 397 MB | 2.67x | 53.1 MB (pg_dump -Fc) |
+| RDF triplestore | 1.74 GB | 11.70x | 126 MB (nt.gz) |
+
+Median wall time on the 150 MB corpus, warm cache (peak RAM / CPU / read
+bytes: [out/report/tables/t2_measured_medians.csv](out/report/tables/t2_measured_medians.csv)):
+
+| Format | Q1 | Q2 | Q3 | Q4 | Q5 | Q6 | Q7 |
+|---|---|---|---|---|---|---|---|
+| CSV | 0.06 s | 0.17 s | 1.15 s | 0.72 s | 1.09 s | 0.28 s | 1.27 s |
+| JSON-LD | 0.11 s | 1.48 s | 9.9 s | 9.9 s | 10 s | 9.8 s | 10 s |
+| SQLite | 0.06 s | 0.06 s | 0.11 s | 0.06 s | 0.77 s | 0.06 s | 0.71 s |
+| PostgreSQL | 0.22 s | 0.22 s | 0.22 s | 0.22 s | 0.44 s | 0.22 s | 0.33 s |
+| RDF triplestore | 0.11 s | 2.81 s | 46 s | 20 s | 60 s | 0.17 s | 41 s |
+
+### Projections (task 10)
+
+Projected wall times at 600 GB - `(!)` = IMPRACTICAL (over 1 hour),
+`FAIL` = over 24 hours; 1.2 TB and 6 TB scales:
+[out/report/tables/t3_projected_600gb.csv](out/report/tables/t3_projected_600gb.csv)
+and `out/bench/extrapolation.csv`:
+
+| Format | Q1 | Q2 | Q3 | Q4 | Q5 | Q6 | Q7 |
+|---|---|---|---|---|---|---|---|
+| CSV | 0.06 s | 7.1 min | 1.2 h (!) | 44 min | 1.2 h (!) | 15 min | 1.4 h (!) |
+| JSON-LD | 0.11 s | 1.5 h (!) | 11 h (!) | 11 h (!) | 11 h (!) | 10.9 h (!) | 11.2 h (!) |
+| SQLite | 0.06 s | 1.3 s | 5.7 min | 0.06 s | 47 min | 5.2 s | 44 min |
+| PostgreSQL | 0.22 s | 0.22 s | 3.4 s | 11 s | 3.7 min | 4.1 s | 1.9 min |
+| RDF triplestore | 0.11 s | 3.0 h (!) | 2.1 d FAIL | 22 h (!) | 2.8 d FAIL | 5.9 min | 1.9 d FAIL |
+
+Hardware sizing at 600 GB - bill of materials with street prices as of
+2026-07-11 (component sources: REPORT.md section 7 and
+[out/config/hw_prices.yaml](out/config/hw_prices.yaml)):
+
+| Format | Configuration | Total |
+|---|---|---|
+| CSV, JSON-LD, SQLite, PostgreSQL (each) | 1 x 1U Supermicro AS-1015CS-TNR (16 cores); 1 x 64 GB DDR5 ECC RDIMM ($1,887); 1 x 7.68 TB NVMe U.2 ($3,995) | $10,500 |
+| RDF triplestore | 2 nodes; 22 x 64 GB DDR5 ECC RDIMM; 2 x 7.68 TB NVMe U.2 | $58,740 |
+
+At 6 TB PostgreSQL still fits one 16-core node - 9 RDIMM modules, 3 NVMe
+drives, $33,586; the RDF triplestore needs 14 nodes and $527,732
+(projected; full table: `out/bench/hardware_sizing.csv`).
+
+### Charts (task 11)
+
+![C1 - disk footprint per format, log scale](out/report/charts/c1_disk_footprint.png)
+![C2 - RAM requirement per format, log scale](out/report/charts/c2_ram_requirement.png)
+![C3 - per-query wall time, one chart per query](out/report/charts/c3_q1_wall_time.png)
+![C4 - degradation curves 150 MB to 6 TB, log-log](out/report/charts/c4_degradation_full_scan.png)
+![C5 - weighted score, stacked bars](out/report/charts/c5_weighted_score.png)
+![C6 - cost vs performance at 600 GB, log-log](out/report/charts/c6_cost_vs_performance.png)
+
+(Task 11 writes the charts to `out/report/charts/` under exactly these
+names; c3 covers all seven queries - `c3_q1..q7_wall_time.png` - and c4 has
+three query classes: `point_read`, `indexed`, `full_scan`.)
+
+### Final scoreboard (task 11)
+
+Weighted scoring, max 80 points: search speed x5 (0-50), RAM economy x2
+(0-20), disk economy x1 (0-10); a FAIL projection at 600 GB zeroes the
+search subscore. Values below are the projected-600 GB scoreboard from the
+2026-07-11 pipeline run (seed 20260711); the generated source of truth is
+`out/report/tables/t5_weighted_scores.csv`, and REPORT.md also carries the
+measured-scale scoreboard.
+
+| Format | Search (0-50) | RAM (0-20) | Disk (0-10) | Total (0-80) |
+|---|---|---|---|---|
+| CSV | 0.6 | 20.0 | 8.3 | 28.9 |
+| JSON-LD | 0.1 | 20.0 | 4.0 | 24.0 |
+| SQLite | 21.8 | 15.2 | 10.0 | 47.0 |
+| PostgreSQL | 50.0 | 5.5 | 3.1 | 58.6 |
+| RDF | 0.0 | 0.2 | 0.7 | 0.9 |
+
+| Use case | Recommended format |
+|---|---|
+| Analysis | PostgreSQL (SQLite acceptable single-user up to ~600 GB) |
+| Report generation | PostgreSQL (track_summary materialized view) |
+| Search / filtering | PostgreSQL or SQLite (indexed) |
+| Archiving | CSV tree + tar.gz, plus pg_dump of the system of record |
+| Inter-lab exchange | Hybrid JSON-LD (metadata + sourceFile links to CSV) |
+
+## Conventions
+
+- IDs: `batch:B721`, `wafer:B721-W1`, `coupon:B721-W1-C07`,
+  `track:B721-W1-C07-T2`, runs `R1..R4`.
+- Timestamps: ISO-8601 UTC; all corpus timestamps are simulated.
+- Units are explicit ASCII suffixes in column names (`_mN`, `_um`, `_GPa`,
+  `_wtpct`, `_nm`).
+- Random seed `20260711`, read from `lab_config.yaml`; regeneration is
+  byte-identical (proof: `MANIFEST.csv` sha256).
+- Python: tabs for indentation; closed dependency list; streaming discipline
+  (the corpus never fits in memory by design).

bench_runner.py

@@ -0,0 +1,527 @@
+"""Task 09 - Benchmark execution and resource measurement.
+
+Runs the 35-cell matrix (Q1-Q7 x 5 formats) in randomized order. Every
+cell: 1 warm-up run (recorded as run 0, excluded from statistics) + 3
+measured runs, each in its OWN subprocess sampled with psutil every 50 ms
+for peak RSS / CPU time / read bytes. Result rows of every run are
+validated against out/bench/expected/ (tolerance 1e-9) and recorded as
+checksum_ok. Writes results_raw.csv, results_median.csv, PG EXPLAIN plans,
+PG host/server metrics, the archival storage sizes, an environment record,
+and the 09.ok marker. Spec: docs/specs/09_benchmark_execution.md; protocol:
+docs/rules/bench-methodology.md.
+
+Environment adaptations (bench-methodology section 6):
+- the cold-cache matrix pass is SKIPPED (Windows host, no page-cache drop);
+- PostgreSQL runs on a remote host: no backend-PID psutil sampling;
+  instead best-effort Prometheus host CPU/RAM windows per PG cell and per
+  session, plus best-effort pg_stat_statements per-query session deltas;
+- on Windows the 50 ms sample reads the OS-maintained peak working set
+  (peak_wset), so sub-interval children still report a true peak RSS.
+
+After a run times out, the cell's remaining runs are SKIPPED (never retry
+a timeout - bench-methodology section 3); only executed runs appear in the
+raw results and the cell reports TIMEOUT.
+"""
+
+from __future__ import annotations
+
+import csv
+import gzip
+import logging
+import random
+import re
+import shutil
+import statistics
+import subprocess
+import sys
+import tarfile
+import time
+from dataclasses import dataclass, field
+from pathlib import Path
+
+import psutil
+import psycopg
+
+from common.monitoring import get_prom_url, host_window_metrics
+from common.pg import get_dsn
+from common.pipeline import (
+	REPO_ROOT,
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	remove_stale_marker,
+	write_marker,
+)
+from common.results import compare, parse_rows
+from common.storage_sizes import update_storage_sizes
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "09"
+DEPENDS_ON = ["08"]
+
+CELL_TIMEOUT_S = 30 * 60   # hard per-run timeout, spec 09
+SAMPLE_INTERVAL_S = 0.05   # psutil sampling cadence, spec 09
+MEASURED_RUNS = 3          # 1 warm-up + 3 measured per cell, spec 09
+
+FORMATS = ("csv", "json", "sqlite", "pg", "rdf")
+QUERY_NUMBERS = tuple(range(1, 8))
+
+RAW_HEADER = [
+	"format", "query", "run", "wall_s", "peak_rss_mb", "cpu_s",
+	"cpu_util_pct", "read_mb", "rows", "checksum_ok", "status", "exec_order",
+]
+MEDIAN_HEADER = [
+	"format", "query", "wall_s_median", "wall_s_min", "wall_s_max",
+	"peak_rss_mb_median", "cpu_s_median", "cpu_util_pct_median",
+	"read_mb_median", "rows", "checksum_ok", "status",
+]
+
+# storage_sizes.csv completeness gate: every (format, variant) that must
+# exist after consolidation (working + archival footprints, spec 09)
+REQUIRED_SIZE_ROWS = {
+	("csv", "tree"), ("csv", "targz"),
+	("json", "full"), ("json", "hybrid"), ("json", "targz"),
+	("sqlite", "db"), ("sqlite", "db_gz"),
+	("pg", "live"), ("pg", "dump"),
+	("rdf", "nt_gz"), ("rdf", "ttl"), ("rdf", "store"),
+}
+
+
+class CellTimeoutError(Exception):
+	pass
+
+
+@dataclass
+class RunResult:
+	status: str
+	wall_s: float = 0.0
+	peak_rss_mb: float = 0.0
+	cpu_s: float = 0.0
+	cpu_util_pct: float = 0.0
+	read_mb: float = 0.0
+	rows: int | None = None
+	checksum_ok: bool | None = None
+
+	def raw_row(self, fmt: str, qn: int, run: int, order: int) -> list:
+		ok = "" if self.checksum_ok is None else str(self.checksum_ok).lower()
+		return [
+			fmt, f"q{qn}", run, round(self.wall_s, 4), round(self.peak_rss_mb, 1),
+			round(self.cpu_s, 3), round(self.cpu_util_pct, 1), round(self.read_mb, 2),
+			"" if self.rows is None else self.rows, ok, self.status, order,
+		]
+
+
+def _kill_tree(proc: subprocess.Popen) -> None:
+	"""Kill the child and every descendant (the stub's real interpreter)."""
+	try:
+		for child in psutil.Process(proc.pid).children(recursive=True):
+			try:
+				child.kill()
+			except psutil.NoSuchProcess:
+				logger.debug("kill_tree: pid %d already exited", child.pid)
+	except psutil.Error:
+		logger.warning("kill_tree: descendants of pid %d not enumerable", proc.pid, exc_info=True)
+	proc.kill()
+	proc.wait()
+
+
+def _run_child(cmd: list[str], stdout_path: Path, stderr_path: Path) -> tuple[int, RunResult]:
+	"""Spawn one query subprocess and sample its process TREE until exit.
+
+	The venv python.exe on Windows is a launcher stub: the real interpreter
+	runs as its child, so sampling only the spawned PID measures the stub
+	(~4 MB flat, ~0 CPU). Every sample therefore walks the descendant tree
+	and keeps per-PID last-known CPU / read bytes and the OS peak working
+	set; totals are sums over all PIDs ever seen. The flat stub adds ~4 MB
+	to the peak-RSS sum - negligible against real query footprints.
+	"""
+	cores = psutil.cpu_count(logical=True)
+	seen: dict[int, list] = {}  # pid -> [cpu_s, read_bytes, peak_rss_bytes]
+	timed_out = False
+	start = time.perf_counter()
+	with open(stdout_path, "wb") as out_fh, open(stderr_path, "wb") as err_fh:
+		proc = subprocess.Popen(cmd, stdout=out_fh, stderr=err_fh, cwd=REPO_ROOT)
+		try:
+			try:
+				root = psutil.Process(proc.pid)
+			except psutil.Error:
+				root = None  # child already gone: exited faster than one sample
+			while root is not None and proc.poll() is None:
+				try:
+					procs = [root, *root.children(recursive=True)]
+				except psutil.Error:
+					break
+				for p in procs:
+					try:
+						mem = p.memory_info()
+						cpu = p.cpu_times()
+						io = p.io_counters()
+					except psutil.Error:
+						continue  # that process exited between listing and sampling
+					entry = seen.setdefault(p.pid, [0.0, 0, 0])
+					entry[0] = cpu.user + cpu.system
+					entry[1] = io.read_bytes
+					entry[2] = max(entry[2], getattr(mem, "peak_wset", 0) or mem.rss)
+				if time.perf_counter() - start > CELL_TIMEOUT_S:
+					timed_out = True
+					break
+				time.sleep(SAMPLE_INTERVAL_S)
+			if timed_out:
+				_kill_tree(proc)
+				raise CellTimeoutError(f"exceeded {CELL_TIMEOUT_S} s")
+			returncode = proc.wait()
+		finally:
+			if proc.poll() is None:
+				_kill_tree(proc)
+	wall = time.perf_counter() - start
+	cpu_s = sum(entry[0] for entry in seen.values())
+	result = RunResult(
+		status="", wall_s=wall,
+		peak_rss_mb=sum(entry[2] for entry in seen.values()) / 1048576,
+		cpu_s=cpu_s,
+		cpu_util_pct=100.0 * cpu_s / wall / cores if wall > 0 else 0.0,
+		read_mb=sum(entry[1] for entry in seen.values()) / 1048576,
+	)
+	return returncode, result
+
+
+def _run_cell_once(fmt: str, qn: int, cmd: list[str], tmp_dir: Path, expected: list[tuple]) -> RunResult:
+	"""One benchmark run with pattern C isolation (code-error-handling)."""
+	stdout_path = tmp_dir / "stdout.txt"
+	stderr_path = tmp_dir / "stderr.txt"
+	try:
+		returncode, result = _run_child(cmd, stdout_path, stderr_path)
+	except CellTimeoutError:
+		logger.error("cell %s/q%d timed out after %d s", fmt, qn, CELL_TIMEOUT_S)
+		return RunResult(status="TIMEOUT", wall_s=CELL_TIMEOUT_S)
+	except Exception:
+		logger.error("cell %s/q%d failed to execute", fmt, qn, exc_info=True)
+		return RunResult(status="ERROR")
+	if returncode != 0:
+		tail = stderr_path.read_text(encoding="utf-8", errors="replace")[-2000:]
+		logger.error("cell %s/q%d exited %d; stderr tail: %s", fmt, qn, returncode, tail)
+		result.status = "ERROR"
+		return result
+	actual = parse_rows(stdout_path.read_text(encoding="ascii"))
+	diff = compare(expected, actual)
+	if diff:
+		logger.error("cell %s/q%d checksum mismatch: %s", fmt, qn, diff)
+	result.status = "OK"
+	result.rows = len(actual)
+	result.checksum_ok = diff is None
+	return result
+
+
+@dataclass
+class CellOutcome:
+	fmt: str
+	qn: int
+	measured: list[RunResult] = field(default_factory=list)
+	timed_out: bool = False  # any run incl. warm-up: remaining runs skipped
+	window: tuple[float, float] | None = None
+
+	def status(self) -> str:
+		if self.timed_out or any(r.status == "TIMEOUT" for r in self.measured):
+			return "TIMEOUT"
+		if len(self.measured) < MEASURED_RUNS or any(r.status == "ERROR" for r in self.measured):
+			return "ERROR"
+		if not all(r.checksum_ok for r in self.measured):
+			return "INVALID"
+		return "OK"
+
+
+def _median_row(outcome: CellOutcome) -> list:
+	status = outcome.status()
+	fmt, qn, measured = outcome.fmt, outcome.qn, outcome.measured
+	if status != "OK":
+		ok = "false" if status == "INVALID" else ""
+		return [fmt, f"q{qn}", "", "", "", "", "", "", "", "", ok, status]
+	walls = [r.wall_s for r in measured]
+	return [
+		fmt, f"q{qn}",
+		round(statistics.median(walls), 4), round(min(walls), 4), round(max(walls), 4),
+		round(statistics.median(r.peak_rss_mb for r in measured), 1),
+		round(statistics.median(r.cpu_s for r in measured), 3),
+		round(statistics.median(r.cpu_util_pct for r in measured), 1),
+		round(statistics.median(r.read_mb for r in measured), 2),
+		measured[0].rows, "true", status,
+	]
+
+
+def _capture_explain(dsn: str, queries_dir: Path, explain_dir: Path) -> None:
+	"""EXPLAIN (ANALYZE, BUFFERS) per PG query; verify Q5 avoids track_summary."""
+	explain_dir.mkdir(parents=True, exist_ok=True)
+	with psycopg.connect(dsn) as conn:
+		for qn in QUERY_NUMBERS:
+			sql = (queries_dir / "pg" / f"q{qn}.sql").read_text(encoding="ascii")
+			plan = "\n".join(r[0] for r in conn.execute("EXPLAIN (ANALYZE, BUFFERS) " + sql).fetchall())
+			(explain_dir / f"q{qn}.txt").write_text(plan + "\n", encoding="utf-8", newline="\n")
+			if qn == 5 and "track_summary" in plan:
+				raise RuntimeError("q5 plan touches track_summary - the forced full scan is violated (bench-methodology section 4)")
+	logger.info("EXPLAIN plans captured to %s", explain_dir)
+
+
+PG_STAT_QUERY = """
+SELECT query, calls, total_exec_time, rows,
+       shared_blks_hit, shared_blks_read, temp_blks_read, temp_blks_written
+FROM pg_stat_statements
+"""
+PG_STAT_COLUMNS = [
+	"calls", "total_exec_time_ms", "rows",
+	"shared_blks_hit", "shared_blks_read", "temp_blks_read", "temp_blks_written",
+]
+_QTAG = re.compile(r"^-- Q(\d):")
+
+
+def _pg_statements_snapshot(dsn: str) -> dict[int, list[float]] | None:
+	"""Sum pg_stat_statements counters per benchmark query (matched by the
+	leading '-- Q<N>:' comment). Best-effort: None when the extension is
+	unavailable - server-side stats are auxiliary and never fail the task.
+	"""
+	try:
+		sums: dict[int, list[float]] = {}
+		with psycopg.connect(dsn) as conn:
+			for query_text, *values in conn.execute(PG_STAT_QUERY):
+				match = _QTAG.match(query_text)
+				if not match:
+					continue
+				acc = sums.setdefault(int(match.group(1)), [0.0] * len(values))
+				for i, v in enumerate(values):
+					acc[i] += float(v)
+		return sums
+	except Exception:
+		logger.warning("pg_stat_statements snapshot skipped (extension unavailable or not readable)", exc_info=True)
+		return None
+
+
+def _write_pg_server_stats(path: Path, start: dict[int, list[float]], end: dict[int, list[float]]) -> None:
+	with open(path, "w", encoding="ascii", newline="") as fh:
+		writer = csv.writer(fh)
+		writer.writerow(["query"] + PG_STAT_COLUMNS)
+		for qn in QUERY_NUMBERS:
+			before = start.get(qn, [0.0] * len(PG_STAT_COLUMNS))
+			after = end.get(qn)
+			if after is None:
+				logger.warning("pg_stat_statements: q%d missing in the end snapshot (entry evicted?)", qn)
+				writer.writerow([f"q{qn}"] + [""] * len(PG_STAT_COLUMNS))
+				continue
+			delta = [after[i] - before[i] for i in range(len(PG_STAT_COLUMNS))]
+			writer.writerow([f"q{qn}"] + [round(v, 3) for v in delta])
+	logger.info("pg server-side statement deltas written to %s", path)
+
+
+def _tree_bytes(root: Path) -> int:
+	return sum(p.stat().st_size for p in root.rglob("*") if p.is_file())
+
+
+def _consolidate_storage_sizes(out_root: Path, bench_dir: Path) -> None:
+	"""Build archival variants, record all sizes, verify completeness."""
+	archives = bench_dir / "archives"
+	archives.mkdir(parents=True, exist_ok=True)
+
+	csv_tgz = archives / "csv_tree.tar.gz"
+	logger.info("archiving out/csv -> %s", csv_tgz)
+	with tarfile.open(csv_tgz, "w:gz") as tar:
+		tar.add(out_root / "csv", arcname="csv")
+	update_storage_sizes(bench_dir, "csv", [
+		("tree", _tree_bytes(out_root / "csv")), ("targz", csv_tgz.stat().st_size),
+	])
+
+	json_tgz = archives / "json.tar.gz"
+	logger.info("archiving out/json -> %s", json_tgz)
+	with tarfile.open(json_tgz, "w:gz") as tar:
+		tar.add(out_root / "json", arcname="json")
+	update_storage_sizes(bench_dir, "json", [("targz", json_tgz.stat().st_size)])
+
+	db_gz = archives / "tribo.db.gz"
+	logger.info("compressing out/sqlite/tribo.db -> %s", db_gz)
+	with open(out_root / "sqlite" / "tribo.db", "rb") as src, gzip.open(db_gz, "wb") as dst:
+		shutil.copyfileobj(src, dst, 1048576)
+	update_storage_sizes(bench_dir, "sqlite", [("db_gz", db_gz.stat().st_size)])
+
+	# pg_dump -Fc is internally gzip-compressed: it IS the pg archival variant
+	update_storage_sizes(bench_dir, "pg", [("dump", (out_root / "pg" / "tribo.dump").stat().st_size)])
+
+	present = set()
+	with open(bench_dir / "storage_sizes.csv", encoding="ascii") as fh:
+		for row in csv.reader(fh):
+			present.add((row[0], row[1]))
+	missing = REQUIRED_SIZE_ROWS - present
+	if missing:
+		raise ValueError(f"storage_sizes.csv incomplete after consolidation: missing {sorted(missing)}")
+	logger.info("storage_sizes.csv consolidated: all %d required rows present", len(REQUIRED_SIZE_ROWS))
+
+
+def _environment_record(path: Path, dsn: str) -> None:
+	import platform
+
+	lines = [
+		f"benchmark_host_os: {platform.platform()}",
+		f"benchmark_host_cpu: {platform.processor()}",
+		f"benchmark_host_cores_logical: {psutil.cpu_count(logical=True)}",
+		f"benchmark_host_cores_physical: {psutil.cpu_count(logical=False)}",
+		f"benchmark_host_ram_gib: {round(psutil.virtual_memory().total / 1073741824, 1)}",
+		f"python: {platform.python_version()}",
+	]
+	try:
+		disks = subprocess.run(
+			["powershell", "-NoProfile", "-Command",
+				"Get-PhysicalDisk | ForEach-Object { $_.FriendlyName + ' / ' + $_.MediaType + ' / ' + [math]::Round($_.Size / 1GB) + ' GB' }"],
+			capture_output=True, text=True, timeout=60, check=True)
+		lines.append("benchmark_host_disks: " + "; ".join(ln.strip() for ln in disks.stdout.splitlines() if ln.strip()))
+	except Exception:
+		logger.warning("disk info skipped (Get-PhysicalDisk query failed - non-Windows host or no permission)", exc_info=True)
+	import sqlite3
+
+	import ijson
+	import pyoxigraph
+
+	lines += [
+		f"sqlite: {sqlite3.sqlite_version}",
+		f"psycopg: {psycopg.__version__}",
+		f"pyoxigraph: {pyoxigraph.__version__}",
+		f"ijson: {ijson.__version__}",
+		f"psutil: {psutil.__version__}",
+	]
+	with psycopg.connect(dsn) as conn:
+		pg_version = conn.execute("SELECT version()").fetchone()[0]
+		lines.append(f"postgresql_server: {pg_version}")
+		lines.append(f"postgresql_host: {conn.info.host} (remote; 4 cores / 3.8 GiB, see .done/06.ok)")
+	lines += [
+		"cold_cache_pass: skipped (Windows host, no page-cache drop - bench-methodology section 6)",
+		"pg_backend_sampling: not possible for a remote server; Prometheus windows + pg_stat_statements deltas instead",
+	]
+	path.write_text("\n".join(lines) + "\n", encoding="utf-8", newline="\n")
+	logger.info("environment record written to %s", path)
+
+
+def _run_matrix(bench_dir: Path, expected: dict[int, list[tuple]], seed: int) -> list[CellOutcome]:
+	cells = [(fmt, qn) for fmt in FORMATS for qn in QUERY_NUMBERS]
+	random.Random(f"{seed}-bench-order").shuffle(cells)
+	tmp_dir = bench_dir / "tmp"
+	tmp_dir.mkdir(parents=True, exist_ok=True)
+	outcomes: list[CellOutcome] = []
+	with open(bench_dir / "results_raw.csv", "w", encoding="ascii", newline="") as raw_fh:
+		writer = csv.writer(raw_fh)
+		writer.writerow(RAW_HEADER)
+		for order, (fmt, qn) in enumerate(cells, start=1):
+			cmd = [sys.executable, str(bench_dir / "queries" / fmt / f"q{qn}.py")]
+			outcome = CellOutcome(fmt, qn)
+			window_start = time.time()
+			for run in range(MEASURED_RUNS + 1):  # run 0 = warm-up
+				result = _run_cell_once(fmt, qn, cmd, tmp_dir, expected[qn])
+				writer.writerow(result.raw_row(fmt, qn, run, order))
+				raw_fh.flush()
+				if run > 0:
+					outcome.measured.append(result)
+				logger.info("cell %s/q%d run %d: %s wall=%.3fs rss=%.1fMB", fmt, qn, run, result.status, result.wall_s, result.peak_rss_mb)
+				if result.status == "TIMEOUT":
+					outcome.timed_out = True
+					logger.error("cell %s/q%d: skipping remaining runs after timeout", fmt, qn)
+					break
+			outcome.window = (window_start, time.time())
+			outcomes.append(outcome)
+			status = outcome.status()
+			walls = [r.wall_s for r in outcome.measured if r.status == "OK"]
+			detail = f"median {statistics.median(walls):.3f} s" if status == "OK" else status
+			print(f"[{order}/{len(cells)}] {fmt}/q{qn}: {detail}")
+	shutil.rmtree(tmp_dir)
+	return outcomes
+
+
+def _write_pg_host_metrics(path: Path, prom_url: str, outcomes: list[CellOutcome]) -> None:
+	with open(path, "w", encoding="ascii", newline="") as fh:
+		writer = csv.writer(fh)
+		writer.writerow([
+			"query", "window_s", "host_cpu_avg_pct", "host_cpu_max_pct",
+			"host_ram_used_baseline_mb", "host_ram_used_peak_mb", "host_ram_used_delta_mb",
+		])
+		for outcome in sorted((o for o in outcomes if o.fmt == "pg"), key=lambda o: o.qn):
+			start, end = outcome.window
+			metrics = host_window_metrics(prom_url, start, end)
+			row = [f"q{outcome.qn}", round(end - start, 1)]
+			if metrics is None:
+				row += [""] * 5
+			else:
+				row += [
+					metrics["host_cpu_avg_pct"], metrics["host_cpu_max_pct"],
+					metrics["host_ram_used_baseline_mb"], metrics["host_ram_used_peak_mb"],
+					metrics["host_ram_used_delta_mb"],
+				]
+			writer.writerow(row)
+	logger.info("per-cell PG host metrics written to %s", path)
+
+
+def main() -> int:
+	args = parse_task_args("Task 09: execute the Q1-Q7 x 5 formats benchmark matrix")
+	out_root: Path = args.out_root
+	bench_dir = out_root / "bench"
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		dump_path = out_root / "pg" / "tribo.dump"
+		if not dump_path.exists():
+			raise FileNotFoundError(f"{dump_path} missing - produce it with pg_dump -Fc before task 09 (storage consolidation needs it)")
+		dsn = get_dsn()
+		expected = {qn: parse_rows((bench_dir / "expected" / f"q{qn}.rows.csv").read_text(encoding="ascii")) for qn in QUERY_NUMBERS}
+		seed = load_lab_config(out_root)["seed"]
+		logger.warning("cold-cache pass skipped (Windows host - bench-methodology section 6)")
+		_environment_record(bench_dir / "environment.txt", dsn)
+
+		prom_url = get_prom_url()
+		stats_start = _pg_statements_snapshot(dsn)
+		session_start = time.time()
+		outcomes = _run_matrix(bench_dir, expected, seed)
+		session_end = time.time()
+
+		with open(bench_dir / "results_median.csv", "w", encoding="ascii", newline="") as fh:
+			writer = csv.writer(fh)
+			writer.writerow(MEDIAN_HEADER)
+			for outcome in sorted(outcomes, key=lambda o: (o.fmt, o.qn)):
+				writer.writerow(_median_row(outcome))
+
+		# snapshot BEFORE the EXPLAIN pass so its executions cannot pollute the deltas
+		if stats_start is not None:
+			stats_end = _pg_statements_snapshot(dsn)
+			if stats_end is not None:
+				_write_pg_server_stats(bench_dir / "pg_server_stats.csv", stats_start, stats_end)
+		_capture_explain(dsn, bench_dir / "queries", bench_dir / "explain")
+		session_host = None
+		if prom_url:
+			_write_pg_host_metrics(bench_dir / "pg_host_metrics.csv", prom_url, outcomes)
+			session_host = host_window_metrics(prom_url, session_start, session_end)
+		else:
+			logger.warning("TRIBO_PROM_URL unset - PG host metrics skipped")
+
+		_consolidate_storage_sizes(out_root, bench_dir)
+
+		by_status: dict[str, int] = {}
+		for outcome in outcomes:
+			by_status[outcome.status()] = by_status.get(outcome.status(), 0) + 1
+		entries = {
+			"cells_total": len(outcomes),
+			"cells_ok": by_status.get("OK", 0),
+			"cells_invalid": by_status.get("INVALID", 0),
+			"cells_timeout": by_status.get("TIMEOUT", 0),
+			"cells_error": by_status.get("ERROR", 0),
+			"session_wall_s": round(session_end - session_start, 1),
+			"cold_cache_pass": "skipped (Windows host)",
+		}
+		if session_host:
+			entries.update({f"session_{k}": v for k, v in session_host.items()})
+		marker_path = write_marker(out_root, TASK_ID, entries)
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	print(f"task 09 ok: {entries['cells_ok']}/{entries['cells_total']} cells OK "
+		f"(invalid={entries['cells_invalid']}, timeout={entries['cells_timeout']}, error={entries['cells_error']}), "
+		f"session {entries['session_wall_s']} s")
+	print(f"results: {bench_dir / 'results_raw.csv'}, {bench_dir / 'results_median.csv'}")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

common/__init__.py

@@ -0,0 +1 @@
+"""Shared helpers for the pipeline tasks (see docs/rules/build-pipeline-tasks.md)."""

common/corpus.py

@@ -0,0 +1,73 @@
+"""Reading the CSV corpus with MANIFEST validation (converters, tasks 04-07).
+
+Every file consumed from ./out/csv/ is verified against MANIFEST.csv (byte
+size, sha256, row count) before its content is trusted - a mismatch aborts
+the converter (docs/rules/test-pipeline-validation.md section 2).
+"""
+
+from __future__ import annotations
+
+import hashlib
+import logging
+import re
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+MANIFEST_HEADER = "path,rows,bytes,sha256"
+
+INT_RE = re.compile(r"^-?[0-9]+$")
+FLOAT_RE = re.compile(r"^-?([0-9]+\.[0-9]*|\.[0-9]+|[0-9]+)([eE][+-]?[0-9]+)?$")
+
+
+class ValidationError(RuntimeError):
+	pass
+
+
+def coerce(s: str) -> int | float | str:
+	if INT_RE.match(s):
+		return int(s)
+	if FLOAT_RE.match(s):
+		return float(s)
+	return s
+
+
+class CorpusReader:
+	def __init__(self, csv_root: Path) -> None:
+		self.csv_root = csv_root
+		self.manifest = self._load_manifest()
+
+	def _load_manifest(self) -> dict[str, tuple[int, int, str]]:
+		manifest: dict[str, tuple[int, int, str]] = {}
+		with open(self.csv_root / "MANIFEST.csv", encoding="ascii") as fh:
+			header = fh.readline().strip()
+			if header != MANIFEST_HEADER:
+				raise ValidationError(f"MANIFEST.csv: unexpected header {header!r}")
+			for line in fh:
+				path, rows, nbytes, sha = line.strip().split(",")
+				manifest[path] = (int(rows), int(nbytes), sha)
+		logger.info("manifest loaded: %d entries", len(manifest))
+		return manifest
+
+	def read(self, relpath: str) -> tuple[list[str], list[list[str]]]:
+		"""Return (header, raw string rows), verified against MANIFEST."""
+		data = (self.csv_root / relpath).read_bytes()
+		exp_rows, exp_bytes, exp_sha = self.manifest[relpath]
+		if len(data) != exp_bytes or hashlib.sha256(data).hexdigest() != exp_sha:
+			raise ValidationError(f"{relpath}: bytes/sha256 mismatch vs MANIFEST")
+		lines = data.decode("ascii").split("\n")
+		header = lines[0].split(",")
+		rows = [ln.split(",") for ln in lines[1:] if ln]
+		if len(rows) != exp_rows:
+			raise ValidationError(f"{relpath}: {len(rows)} rows != manifest {exp_rows}")
+		return header, rows
+
+	def dicts(self, relpath: str) -> list[dict]:
+		"""Rows as dicts: lowercased CSV headers as keys, numerics coerced,
+		empty cells omitted (docs/rules/data-naming-units.md section 4)."""
+		header, rows = self.read(relpath)
+		keys = [h.lower() for h in header]
+		return [
+			{k: coerce(v) for k, v in zip(keys, row) if v != ""}
+			for row in rows
+		]

common/monitoring.py

@@ -0,0 +1,66 @@
+"""Remote host metrics via Prometheus / node_exporter (stdlib urllib only).
+
+The PostgreSQL instance is remote, so psutil cannot attribute its CPU/RAM
+(spec 09 assumed a local instance). The database host runs Prometheus;
+its base URL comes from the TRIBO_PROM_URL environment variable
+(e.g. http://192.168.10.73:9090). Unset = monitoring disabled.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import urllib.parse
+import urllib.request
+
+logger = logging.getLogger(__name__)
+
+QUERY_STEP_S = 5
+HTTP_TIMEOUT_S = 15
+# rate() window must span at least two scrapes (15 s default interval)
+CPU_QUERY = '100 * (1 - avg(rate(node_cpu_seconds_total{mode="idle"}[30s])))'
+RAM_QUERY = "node_memory_MemTotal_bytes - node_memory_MemAvailable_bytes"
+
+
+def get_prom_url() -> str:
+	return os.environ.get("TRIBO_PROM_URL", "")
+
+
+def query_range(base_url: str, promql: str, start: float, end: float) -> list[float]:
+	params = urllib.parse.urlencode({"query": promql, "start": start, "end": end, "step": QUERY_STEP_S})
+	url = base_url.rstrip("/") + "/api/v1/query_range?" + params
+	with urllib.request.urlopen(url, timeout=HTTP_TIMEOUT_S) as resp:
+		data = json.load(resp)
+	if data.get("status") != "success":
+		raise RuntimeError(f"prometheus query failed: {data.get('errorType')}: {data.get('error')}")
+	result = data["data"]["result"]
+	if not result:
+		return []
+	return [float(v) for _, v in result[0]["values"]]
+
+
+def host_window_metrics(base_url: str, start: float, end: float) -> dict | None:
+	"""CPU/RAM of the monitored host over [start, end] wall-clock seconds.
+
+	Returns None (with a warning) on any failure - monitoring is best-effort
+	and must never fail the owning task.
+	"""
+	try:
+		if end - start < 2 * QUERY_STEP_S:
+			end = start + 2 * QUERY_STEP_S
+		cpu = query_range(base_url, CPU_QUERY, start, end)
+		ram = query_range(base_url, RAM_QUERY, start, end)
+		if not cpu or not ram:
+			logger.warning("prometheus returned no samples for the window (expected condition: short window or scrape lag)")
+			return None
+		return {
+			"host_cpu_avg_pct": round(sum(cpu) / len(cpu), 1),
+			"host_cpu_max_pct": round(max(cpu), 1),
+			"host_ram_used_baseline_mb": round(ram[0] / 1048576),
+			"host_ram_used_peak_mb": round(max(ram) / 1048576),
+			"host_ram_used_delta_mb": round((max(ram) - ram[0]) / 1048576),
+		}
+	except Exception:
+		logger.warning("host metrics collection skipped (prometheus unreachable or query failed)", exc_info=True)
+		return None

common/pg.py

@@ -0,0 +1,16 @@
+"""PostgreSQL DSN access (docs/rules/code-config-yaml.md section 3).
+
+The DSN comes from the TRIBO_PG_DSN environment variable and is read ONLY
+here. It may embed a password: never log it, never hardcode it, never
+commit it - log host/database names at most.
+"""
+
+from __future__ import annotations
+
+import os
+
+DEFAULT_DSN = "postgresql://postgres@localhost:5432/tribo"
+
+
+def get_dsn() -> str:
+	return os.environ.get("TRIBO_PG_DSN", DEFAULT_DSN)

common/pipeline.py

@@ -0,0 +1,79 @@
+"""Task-orchestration helpers: CLI shape, dependency markers, completion
+markers, lab config loading (docs/rules/build-pipeline-tasks.md,
+docs/rules/code-python-style.md section 3).
+"""
+
+from __future__ import annotations
+
+import argparse
+import logging
+import sys
+from pathlib import Path
+
+import psutil
+import yaml
+
+logger = logging.getLogger(__name__)
+
+REPO_ROOT = Path(__file__).resolve().parent.parent
+
+
+def parse_task_args(description: str) -> argparse.Namespace:
+	parser = argparse.ArgumentParser(description=description)
+	parser.add_argument("--log-level", default="INFO", choices=["DEBUG", "INFO", "WARNING", "ERROR"])
+	parser.add_argument("--out-root", type=Path, default=REPO_ROOT / "out", help="artifact tree root (default: ./out)")
+	args = parser.parse_args()
+	logging.basicConfig(
+		level=args.log_level,
+		stream=sys.stderr,
+		format="%(asctime)s %(levelname)-7s %(name)s: %(message)s",
+	)
+	return args
+
+
+def check_dependencies(out_root: Path, deps: list[str]) -> None:
+	for dep in deps:
+		marker = out_root / ".done" / f"{dep}.ok"
+		if not marker.exists():
+			raise FileNotFoundError(f"dependency marker missing: {marker} - run task {dep} first")
+
+
+def remove_stale_marker(out_root: Path, task_id: str) -> Path:
+	marker = out_root / ".done" / f"{task_id}.ok"
+	if marker.exists():
+		logger.info("re-run: removing stale marker %s", marker)
+		marker.unlink()
+	return marker
+
+
+def write_marker(out_root: Path, task_id: str, entries: dict) -> Path:
+	marker = out_root / ".done" / f"{task_id}.ok"
+	lines = [f"task: '{task_id}'", "status: ok"] + [f"{k}: {v}" for k, v in entries.items()]
+	marker.parent.mkdir(parents=True, exist_ok=True)
+	with open(marker, "w", encoding="utf-8", newline="\n") as fh:
+		fh.write("\n".join(lines) + "\n")
+	return marker
+
+
+def load_lab_config(out_root: Path) -> dict:
+	with open(out_root / "config" / "lab_config.yaml", encoding="utf-8") as fh:
+		return yaml.safe_load(fh)
+
+
+def process_metrics() -> dict:
+	"""Peak RSS and CPU time of the current process.
+
+	Every converter records its own build cost in its marker so the task 10
+	hardware-sizing model can extrapolate per-format RAM/CPU needs, not just
+	disk. peak_wset exists on Windows; the rss fallback covers Linux, where
+	the value is the current (not peak) RSS - good enough for streaming
+	converters whose RSS is flat by design.
+	"""
+	proc = psutil.Process()
+	mem = proc.memory_info()
+	peak = getattr(mem, "peak_wset", 0) or mem.rss
+	cpu = proc.cpu_times()
+	return {
+		"proc_peak_rss_mb": round(peak / 1048576, 1),
+		"proc_cpu_seconds": round(cpu.user + cpu.system, 1),
+	}

common/queries/__init__.py

@@ -0,0 +1,15 @@
+"""Benchmark query implementations for the file-based formats (spec 08).
+
+One module per format (csv_queries, json_queries, rdf_queries), each with
+functions q1..q7 returning canonical result rows (see common/results.py).
+The runnable wrappers under out/bench/queries/ are generated by
+make_queries.py and delegate here. SQLite / PostgreSQL implementations are
+plain SQL files written by make_queries.py.
+
+The relational formats answer Q2/Q3/Q4/Q7 from the precomputed
+track_summary (their idiomatic strength); the file formats derive the same
+values from raw cycles via common/track_summary.py - identical algorithm,
+identical results (bench-methodology section 5).
+"""
+
+Q1_TRACK_CODE = "B722-W2-C13-T2"  # fixed benchmark track, spec 08

common/queries/csv_queries.py

@@ -0,0 +1,139 @@
+"""Q1-Q7 over the raw CSV tree: pure python + csv module, streaming.
+
+Q1 is a direct path read (the honest CSV strength, spec 08); Q2-Q7 are
+deterministic directory walks with programmatic joins. No manifest
+validation here - queries measure retrieval, not integrity checking.
+"""
+
+from __future__ import annotations
+
+import csv
+from pathlib import Path
+
+from common.queries import Q1_TRACK_CODE
+from common.track_summary import summarize_track
+
+
+def _rows(path: Path) -> tuple[list[str], list[list[str]]]:
+	with open(path, newline="", encoding="ascii") as fh:
+		reader = csv.reader(fh)
+		header = [h.lower() for h in next(reader)]
+		return header, list(reader)
+
+
+def _dict_row(path: Path) -> dict[str, str]:
+	header, rows = _rows(path)
+	return dict(zip(header, rows[0]))
+
+
+def _track_path(csv_root: Path, track_code: str) -> Path:
+	batch, wafer, coupon, track = track_code.split("-")
+	return csv_root / f"batch_{batch}" / f"wafer_{wafer}" / f"coupon_{coupon}" / f"track_{track}"
+
+
+def _coupon_dirs(csv_root: Path):
+	for batch_dir in sorted(csv_root.glob("batch_*")):
+		batch_code = batch_dir.name[6:]
+		for wafer_dir in sorted(batch_dir.glob("wafer_*")):
+			for coupon_dir in sorted(wafer_dir.glob("coupon_*")):
+				yield batch_code, coupon_dir
+
+
+def _track_dirs(coupon_dir: Path) -> list[Path]:
+	return sorted(coupon_dir.glob("track_T*"))
+
+
+def _cofs(track_dir: Path) -> list[float]:
+	_, rows = _rows(track_dir / "cof_vs_cycle.csv")
+	return [float(r[1]) for r in rows]
+
+
+def _mean(values: list[float]) -> float:
+	total = 0.0
+	for v in values:
+		total += v
+	return total / len(values)
+
+
+def q1(csv_root: Path) -> list[tuple]:
+	_, rows = _rows(_track_path(csv_root, Q1_TRACK_CODE) / "cof_vs_cycle.csv")
+	return [(int(r[0]), float(r[1])) for r in rows]
+
+
+def q2(csv_root: Path) -> list[tuple]:
+	out = []
+	for _, coupon_dir in _coupon_dirs(csv_root):
+		info = _dict_row(coupon_dir / "coupon_info.csv")
+		if info["run_code"] == "RESERVE" or not 9.5 <= float(info["au_wtpct_mean"]) <= 10.5:
+			continue
+		tracks = _track_dirs(coupon_dir)
+		env = _dict_row(tracks[0] / "track_info.csv")["environment"]
+		cof_ss = _mean([summarize_track(_cofs(t))[0] for t in tracks])
+		out.append((info["coupon_code"], env, cof_ss))
+	return out
+
+
+def q3(csv_root: Path) -> list[tuple]:
+	out = []
+	for batch_code, coupon_dir in _coupon_dirs(csv_root):
+		info = _dict_row(coupon_dir / "coupon_info.csv")
+		if info["run_code"] == "RESERVE":
+			continue
+		header, rows = _rows(coupon_dir / "nanoindentation.csv")
+		h_col = header.index("hardness_gpa")
+		hardness = _mean([float(r[h_col]) for r in rows])
+		cof_ss = _mean([summarize_track(_cofs(t))[0] for t in _track_dirs(coupon_dir)])
+		out.append((info["coupon_code"], batch_code, hardness, cof_ss))
+	return out
+
+
+def q4(csv_root: Path) -> list[tuple]:
+	out = []
+	for _, coupon_dir in _coupon_dirs(csv_root):
+		for track_dir in _track_dirs(coupon_dir):
+			info = _dict_row(track_dir / "track_info.csv")
+			if info["environment"] != "dry_n2" or float(info["load_mn"]) != 100.0:
+				continue
+			cof_ss = summarize_track(_cofs(track_dir))[0]
+			if cof_ss > 0.20:
+				out.append((info["track_code"], cof_ss))
+	return out
+
+
+def q5(csv_root: Path) -> list[tuple]:
+	sums: dict[str, list[float]] = {}
+	for batch_code, coupon_dir in _coupon_dirs(csv_root):
+		for track_dir in _track_dirs(coupon_dir):
+			run_in = summarize_track(_cofs(track_dir))[2]
+			acc = sums.setdefault(batch_code, [0.0, 0.0])
+			acc[0] += run_in
+			acc[1] += 1
+	return [(batch, acc[0] / acc[1]) for batch, acc in sorted(sums.items())]
+
+
+def q6(csv_root: Path) -> list[tuple]:
+	out = []
+	for _, coupon_dir in _coupon_dirs(csv_root):
+		info = _dict_row(coupon_dir / "coupon_info.csv")
+		if info["run_code"] == "RESERVE":
+			continue
+		tracks = _track_dirs(coupon_dir)
+		load = float(_dict_row(tracks[0] / "track_info.csv")["load_mn"])
+		wear = _mean([float(_dict_row(t / "wear.csv")["wear_volume_um3"]) for t in tracks])
+		out.append((info["coupon_code"], load, wear))
+	return out
+
+
+def q7(csv_root: Path) -> list[tuple]:
+	groups: dict[tuple[str, int], list[float]] = {}
+	for _, coupon_dir in _coupon_dirs(csv_root):
+		for track_dir in _track_dirs(coupon_dir):
+			info = _dict_row(track_dir / "track_info.csv")
+			bucket = int(round(float(info["speed_mm_s"]) * float(info["load_mn"])))
+			cofs = _cofs(track_dir)
+			run_in = summarize_track(cofs)[2]
+			acc = groups.setdefault((info["environment"], bucket), [0.0, 0.0])
+			for cof in cofs[run_in:]:  # cycles strictly past run-in
+				acc[0] += cof
+				acc[1] += 1
+	return [(env, bucket, acc[0] / acc[1]) for (env, bucket), acc in sorted(groups.items())]

common/queries/json_queries.py

@@ -0,0 +1,169 @@
+"""Q1-Q7 over the JSON-LD FULL variant: ijson streaming parser (spec 08).
+
+Each coupon file is streamed event-by-event; scalar coupon fields appear
+before the bulk arrays in the serialization, so filtering queries (Q2) can
+abandon a non-matching file before parsing its megabytes of points.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import ijson
+
+from common.queries import Q1_TRACK_CODE
+from common.track_summary import summarize_track
+
+
+class CouponDoc:
+	__slots__ = ("coupon_code", "batch_code", "run_code", "au", "hardness", "tracks")
+
+	def __init__(self) -> None:
+		self.coupon_code = ""
+		self.batch_code = ""
+		self.run_code = ""
+		self.au = 0.0
+		self.hardness: list[float] = []
+		self.tracks: list[dict] = []
+
+
+def _scan(path: Path, au_range: tuple[float, float] | None = None) -> CouponDoc | None:
+	"""Stream one coupon file; with au_range set, bail out early on mismatch."""
+	doc = CouponDoc()
+	track: dict | None = None
+	with open(path, "rb") as fh:
+		for prefix, event, value in ijson.parse(fh, use_float=True):
+			if prefix == "coupon_code":
+				doc.coupon_code = value
+			elif prefix == "batch_code":
+				doc.batch_code = value
+			elif prefix == "run_code":
+				doc.run_code = value
+				if doc.run_code == "RESERVE":
+					return None
+			elif prefix == "au_wtpct_mean":
+				doc.au = value
+				if au_range and not au_range[0] <= value <= au_range[1]:
+					return None
+			elif prefix == "nanoindentation.indents.item.hardness_gpa":
+				doc.hardness.append(value)
+			elif prefix == "tracks.item" and event == "start_map":
+				track = {"cycles": [], "cofs": []}
+			elif track is not None:
+				if prefix == "tracks.item.track_code":
+					track["track_code"] = value
+				elif prefix == "tracks.item.environment":
+					track["environment"] = value
+				elif prefix == "tracks.item.load_mn":
+					track["load_mn"] = value
+				elif prefix == "tracks.item.speed_mm_s":
+					track["speed_mm_s"] = value
+				elif prefix == "tracks.item.cycles.item.cycle":
+					track["cycles"].append(value)
+				elif prefix == "tracks.item.cycles.item.cof":
+					track["cofs"].append(value)
+				elif prefix == "tracks.item.wear.wear_volume_um3":
+					track["wear"] = value
+				elif prefix == "tracks.item" and event == "end_map":
+					doc.tracks.append(track)
+					track = None
+	return doc
+
+
+def _files(json_root: Path) -> list[Path]:
+	return sorted((json_root / "full").glob("*.jsonld"))
+
+
+def _mean(values: list[float]) -> float:
+	total = 0.0
+	for v in values:
+		total += v
+	return total / len(values)
+
+
+def _cof_ss(doc: CouponDoc) -> float:
+	return _mean([summarize_track(t["cofs"])[0] for t in doc.tracks])
+
+
+def q1(json_root: Path) -> list[tuple]:
+	coupon_code = Q1_TRACK_CODE.rsplit("-", 1)[0]
+	doc = _scan(json_root / "full" / f"{coupon_code}.jsonld")
+	for track in doc.tracks:
+		if track["track_code"] == Q1_TRACK_CODE:
+			return [(int(c), cof) for c, cof in zip(track["cycles"], track["cofs"])]
+	raise LookupError(f"track {Q1_TRACK_CODE} not found")
+
+
+def q2(json_root: Path) -> list[tuple]:
+	out = []
+	for path in _files(json_root):
+		doc = _scan(path, au_range=(9.5, 10.5))
+		if doc is None:
+			continue
+		out.append((doc.coupon_code, doc.tracks[0]["environment"], _cof_ss(doc)))
+	return out
+
+
+def q3(json_root: Path) -> list[tuple]:
+	out = []
+	for path in _files(json_root):
+		doc = _scan(path)
+		if doc is None:
+			continue
+		out.append((doc.coupon_code, doc.batch_code, _mean(doc.hardness), _cof_ss(doc)))
+	return out
+
+
+def q4(json_root: Path) -> list[tuple]:
+	out = []
+	for path in _files(json_root):
+		doc = _scan(path)
+		if doc is None:
+			continue
+		for track in doc.tracks:
+			if track["environment"] != "dry_n2" or track["load_mn"] != 100.0:
+				continue
+			cof_ss = summarize_track(track["cofs"])[0]
+			if cof_ss > 0.20:
+				out.append((track["track_code"], cof_ss))
+	return out
+
+
+def q5(json_root: Path) -> list[tuple]:
+	sums: dict[str, list[float]] = {}
+	for path in _files(json_root):
+		doc = _scan(path)
+		if doc is None:
+			continue
+		acc = sums.setdefault(doc.batch_code, [0.0, 0.0])
+		for track in doc.tracks:
+			acc[0] += summarize_track(track["cofs"])[2]
+			acc[1] += 1
+	return [(batch, acc[0] / acc[1]) for batch, acc in sorted(sums.items())]
+
+
+def q6(json_root: Path) -> list[tuple]:
+	out = []
+	for path in _files(json_root):
+		doc = _scan(path)
+		if doc is None:
+			continue
+		load = doc.tracks[0]["load_mn"]
+		out.append((doc.coupon_code, load, _mean([t["wear"] for t in doc.tracks])))
+	return out
+
+
+def q7(json_root: Path) -> list[tuple]:
+	groups: dict[tuple[str, int], list[float]] = {}
+	for path in _files(json_root):
+		doc = _scan(path)
+		if doc is None:
+			continue
+		for track in doc.tracks:
+			bucket = int(round(track["speed_mm_s"] * track["load_mn"]))
+			run_in = summarize_track(track["cofs"])[2]
+			acc = groups.setdefault((track["environment"], bucket), [0.0, 0.0])
+			for cof in track["cofs"][run_in:]:
+				acc[0] += cof
+				acc[1] += 1
+	return [(env, bucket, acc[0] / acc[1]) for (env, bucket), acc in sorted(groups.items())]

common/queries/rdf_queries.py

@@ -0,0 +1,173 @@
+"""Q1-Q7 over the oxigraph triplestore: SPARQL retrieval (spec 08).
+
+The store holds raw data only (no precomputed summaries), so queries that
+need steady-state COF or run-in retrieve the raw cycles via SPARQL and
+derive the values with the shared algorithm (common/track_summary.py) -
+identical semantics to every other format. The SPARQL text of each query
+is exported to out/bench/queries/rdf/q<N>.sparql by make_queries.py.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+from pyoxigraph import Store
+
+from common.queries import Q1_TRACK_CODE
+from common.track_summary import summarize_track
+
+PREFIX = "PREFIX tribo: <https://sandia.gov/ontology/tribology#>\n"
+TRACK_IRI = f"<https://sandia.gov/ontology/tribology/id/track/{Q1_TRACK_CODE}>"
+
+SPARQL = {
+	"q1": PREFIX + f"""SELECT ?cycle ?cof WHERE {{
+  {TRACK_IRI} tribo:hasCycle ?c .
+  ?c tribo:cycle ?cycle ; tribo:cof ?cof .
+}}""",
+	"q2": PREFIX + """SELECT ?ccode ?env ?track ?cycle ?cof WHERE {
+  ?coupon tribo:au_wtpct_mean ?au ; tribo:coupon_code ?ccode .
+  FILTER(?au >= 9.5 && ?au <= 10.5)
+  ?track tribo:partOf ?coupon ; tribo:environment ?env ; tribo:hasCycle ?c .
+  ?c tribo:cycle ?cycle ; tribo:cof ?cof .
+}""",
+	"q3_hardness": PREFIX + """SELECT ?ccode ?bcode (AVG(?h) AS ?hardness) WHERE {
+  ?coupon tribo:coupon_code ?ccode ; tribo:cutFrom ?wafer ; tribo:nanoindentation ?m .
+  ?wafer tribo:partOf ?batch .
+  ?batch tribo:batch_code ?bcode .
+  ?m tribo:hasIndent ?i .
+  ?i tribo:hardness_gpa ?h .
+} GROUP BY ?ccode ?bcode""",
+	"q3_cycles": PREFIX + """SELECT ?ccode ?track ?cycle ?cof WHERE {
+  ?track a tribo:FrictionTest ; tribo:partOf ?coupon ; tribo:hasCycle ?c .
+  ?coupon tribo:coupon_code ?ccode .
+  ?c tribo:cycle ?cycle ; tribo:cof ?cof .
+}""",
+	"q4": PREFIX + """SELECT ?tcode ?track ?cycle ?cof WHERE {
+  ?track tribo:environment "dry_n2" ; tribo:load_mn ?load ; tribo:track_code ?tcode ; tribo:hasCycle ?c .
+  FILTER(?load = 100)
+  ?c tribo:cycle ?cycle ; tribo:cof ?cof .
+}""",
+	"q5": PREFIX + """SELECT ?bcode ?track ?cycle ?cof WHERE {
+  ?track a tribo:FrictionTest ; tribo:partOf ?coupon ; tribo:hasCycle ?c .
+  ?coupon tribo:cutFrom ?wafer .
+  ?wafer tribo:partOf ?batch .
+  ?batch tribo:batch_code ?bcode .
+  ?c tribo:cycle ?cycle ; tribo:cof ?cof .
+}""",
+	"q6": PREFIX + """SELECT ?ccode ?load (AVG(?wv) AS ?wear) WHERE {
+  ?track tribo:partOf ?coupon ; tribo:load_mn ?load ; tribo:wear ?w .
+  ?w tribo:wear_volume_um3 ?wv .
+  ?coupon tribo:coupon_code ?ccode .
+} GROUP BY ?ccode ?load""",
+	"q7": PREFIX + """SELECT ?env ?speed ?load ?track ?cycle ?cof WHERE {
+  ?track tribo:environment ?env ; tribo:speed_mm_s ?speed ; tribo:load_mn ?load ; tribo:hasCycle ?c .
+  ?c tribo:cycle ?cycle ; tribo:cof ?cof .
+}""",
+}
+
+
+def _store(store_path: Path) -> Store:
+	try:
+		return Store.read_only(str(store_path))
+	except AttributeError:
+		return Store(str(store_path))
+
+
+def _track_cofs(solutions, track_var: str, keep: tuple[str, ...]) -> dict[str, dict]:
+	"""Group cycle solutions by track: ordered cof list + kept literals."""
+	tracks: dict[str, dict] = {}
+	for sol in solutions:
+		key = str(sol[track_var])
+		entry = tracks.get(key)
+		if entry is None:
+			entry = {"pairs": []}
+			for name in keep:
+				entry[name] = sol[name].value
+			tracks[key] = entry
+		entry["pairs"].append((int(sol["cycle"].value), float(sol["cof"].value)))
+	for entry in tracks.values():
+		entry["pairs"].sort()
+		entry["cofs"] = [cof for _, cof in entry["pairs"]]
+	return tracks
+
+
+def _mean(values: list[float]) -> float:
+	total = 0.0
+	for v in values:
+		total += v
+	return total / len(values)
+
+
+def q1(store_path: Path) -> list[tuple]:
+	store = _store(store_path)
+	return [(int(s["cycle"].value), float(s["cof"].value)) for s in store.query(SPARQL["q1"])]
+
+
+def q2(store_path: Path) -> list[tuple]:
+	store = _store(store_path)
+	tracks = _track_cofs(store.query(SPARQL["q2"]), "track", ("ccode", "env"))
+	coupons: dict[tuple[str, str], list[float]] = {}
+	for entry in sorted(tracks.values(), key=lambda e: e["ccode"]):
+		coupons.setdefault((entry["ccode"], entry["env"]), []).append(summarize_track(entry["cofs"])[0])
+	return [(code, env, _mean(values)) for (code, env), values in sorted(coupons.items())]
+
+
+def q3(store_path: Path) -> list[tuple]:
+	store = _store(store_path)
+	hardness = {
+		s["ccode"].value: (s["bcode"].value, float(s["hardness"].value))
+		for s in store.query(SPARQL["q3_hardness"])
+	}
+	tracks = _track_cofs(store.query(SPARQL["q3_cycles"]), "track", ("ccode",))
+	coupons: dict[str, list[float]] = {}
+	for entry in sorted(tracks.values(), key=lambda e: e["ccode"]):
+		coupons.setdefault(entry["ccode"], []).append(summarize_track(entry["cofs"])[0])
+	return [
+		(code, hardness[code][0], hardness[code][1], _mean(values))
+		for code, values in sorted(coupons.items())
+	]
+
+
+def q4(store_path: Path) -> list[tuple]:
+	store = _store(store_path)
+	tracks = _track_cofs(store.query(SPARQL["q4"]), "track", ("tcode",))
+	out = []
+	for entry in sorted(tracks.values(), key=lambda e: e["tcode"]):
+		cof_ss = summarize_track(entry["cofs"])[0]
+		if cof_ss > 0.20:
+			out.append((entry["tcode"], cof_ss))
+	return out
+
+
+def q5(store_path: Path) -> list[tuple]:
+	store = _store(store_path)
+	tracks = _track_cofs(store.query(SPARQL["q5"]), "track", ("bcode",))
+	sums: dict[str, list[float]] = {}
+	for entry in sorted(tracks.values(), key=lambda e: e["bcode"]):
+		acc = sums.setdefault(entry["bcode"], [0.0, 0.0])
+		acc[0] += summarize_track(entry["cofs"])[2]
+		acc[1] += 1
+	return [(batch, acc[0] / acc[1]) for batch, acc in sorted(sums.items())]
+
+
+def q6(store_path: Path) -> list[tuple]:
+	store = _store(store_path)
+	return [
+		(s["ccode"].value, float(s["load"].value), float(s["wear"].value))
+		for s in store.query(SPARQL["q6"])
+	]
+
+
+def q7(store_path: Path) -> list[tuple]:
+	store = _store(store_path)
+	tracks = _track_cofs(store.query(SPARQL["q7"]), "track", ("env", "speed", "load"))
+	groups: dict[tuple[str, int], list[float]] = {}
+	for key in sorted(tracks):
+		entry = tracks[key]
+		bucket = int(round(float(entry["speed"]) * float(entry["load"])))
+		run_in = summarize_track(entry["cofs"])[2]
+		acc = groups.setdefault((entry["env"], bucket), [0.0, 0.0])
+		for cof in entry["cofs"][run_in:]:
+			acc[0] += cof
+			acc[1] += 1
+	return [(env, bucket, acc[0] / acc[1]) for (env, bucket), acc in sorted(groups.items())]

common/relational.py

@@ -0,0 +1,161 @@
+"""Shared relational row stream for the SQLite and PostgreSQL converters.
+
+Walks the CSV corpus once (manifest-validated via CorpusReader) and yields
+(table, row_tuple) pairs in FK-dependency-safe order: a parent row is always
+yielded before any row that references it. Both engines load the SAME
+logical schema from this stream (docs/rules/db-sql-schema.md section 6).
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Iterator
+
+from common.corpus import CorpusReader
+from common.track_summary import summarize_track
+
+logger = logging.getLogger(__name__)
+
+# Dependency-safe order; consumers that buffer rows must flush in this order.
+TABLES = [
+	"instruments",
+	"batches",
+	"runs",
+	"wafers",
+	"coupons",
+	"tracks",
+	"simtra_profiles",
+	"xrf_points",
+	"profilometry_points",
+	"nanoindentation",
+	"afm",
+	"friction_cycles",
+	"friction_loop_points",
+	"wear",
+	"track_summary",
+]
+
+COLUMNS = {
+	"instruments": ("instrument_id", "instrument_code", "name", "role"),
+	"batches": ("batch_id", "batch_code", "pt_gun_tilt_deg", "au_gun_tilt_deg", "pt_power_w", "au_power_w", "pt_discharge_v", "au_discharge_v", "deposition_date"),
+	"runs": ("run_id", "run_code", "environment", "date", "plates", "operator"),
+	"wafers": ("wafer_id", "wafer_code", "batch_id", "wafer_index", "deposition_date", "coupons", "friction_coupons", "reserve_coupons"),
+	"coupons": ("coupon_id", "coupon_code", "wafer_id", "batch_id", "grid_row", "grid_col", "thickness_um", "ra_nm", "au_wtpct_mean", "run_id", "plate", "probe", "square"),
+	"tracks": ("track_id", "track_code", "coupon_id", "run_id", "environment", "load_mn", "stroke_mm", "speed_mm_s", "counterface_id", "started_at"),
+	"simtra_profiles": ("batch_id", "row_no", "angle_deg", "energy_ev", "pt_flux", "au_flux"),
+	"xrf_points": ("coupon_id", "grid_x", "grid_y", "pt_wtpct", "au_wtpct"),
+	"profilometry_points": ("coupon_id", "grid_x", "grid_y", "thickness_um"),
+	"nanoindentation": ("coupon_id", "indent_id", "x_um", "y_um", "hardness_gpa", "reduced_modulus_gpa", "max_load_mn"),
+	"afm": ("coupon_id", "ra_nm", "rq_nm", "image_file"),
+	"friction_cycles": ("track_id", "cycle", "cof"),
+	"friction_loop_points": ("track_id", "cycle", "pt", "position_um", "friction_force_mn"),
+	"wear": ("track_id", "wear_volume_um3", "k_archard", "sliding_distance_m"),
+	"track_summary": ("track_id", "cof_ss_mean", "cof_ss_std", "run_in_cycles"),
+}
+
+EXPECTED_COUNTS_KEYS = TABLES  # every table is count-validated after load
+
+
+def expected_counts(cfg: dict) -> dict[str, int]:
+	v = cfg["volumes"]
+	h = cfg["hierarchy"]
+	return {
+		"instruments": len(cfg["instruments"]),
+		"batches": h["batches"],
+		"runs": cfg["friction_assignment"]["runs"],
+		"wafers": h["batches"] * h["wafers_per_batch"],
+		"coupons": v["coupons_total"],
+		"tracks": v["tracks_total"],
+		"simtra_profiles": h["batches"] * v["simtra_rows_per_batch"],
+		"xrf_points": v["xrf_points_total"],
+		"profilometry_points": v["coupons_total"] * v["profilometry_points_per_coupon"],
+		"nanoindentation": v["coupons_total"] * v["nanoindentation_indents_per_coupon"],
+		"afm": v["coupons_total"],
+		"friction_cycles": v["cycle_rows_total"],
+		"friction_loop_points": v["loop_points_total"],
+		"wear": v["tracks_total"],
+		"track_summary": v["tracks_total"],
+	}
+
+
+def stream_rows(cfg: dict, reader: CorpusReader) -> Iterator[tuple[str, tuple]]:
+	batch_ids: dict[str, int] = {}
+	run_ids: dict[str, int] = {}
+
+	for i, inst in enumerate(cfg["instruments"], 1):
+		yield "instruments", (i, inst["instrument_id"], inst["name"], inst["role"])
+	for i, row in enumerate(reader.dicts("batches.csv"), 1):
+		batch_ids[row["batch_code"]] = i
+		yield "batches", (
+			i, row["batch_code"], row["pt_gun_tilt_deg"], row["au_gun_tilt_deg"],
+			row["pt_power_w"], row["au_power_w"], row["pt_discharge_v"],
+			row["au_discharge_v"], row["deposition_date"],
+		)
+	for i, row in enumerate(reader.dicts("runs.csv"), 1):
+		run_ids[row["run_code"]] = i
+		yield "runs", (i, row["run_code"], row["environment"], row["date"], row["plates"], row["operator"])
+	for code, batch_id in batch_ids.items():
+		for row_no, row in enumerate(reader.dicts(f"simtra/simtra_profile_{code}.csv"), 1):
+			yield "simtra_profiles", (batch_id, row_no, row["angle_deg"], row["energy_ev"], row["pt_flux"], row["au_flux"])
+
+	h = cfg["hierarchy"]
+	fa = cfg["friction_assignment"]
+	wafer_id = coupon_id = track_id = 0
+	for batch in cfg["deposition_matrix"]:
+		b_code = batch["batch_code"]
+		for w in range(1, h["wafers_per_batch"] + 1):
+			wafer_id += 1
+			wafer_dir = f"batch_{b_code}/wafer_W{w}"
+			info = reader.dicts(f"{wafer_dir}/wafer_info.csv")[0]
+			yield "wafers", (
+				wafer_id, info["wafer_code"], batch_ids[b_code], info["wafer_index"],
+				info["deposition_date"], info["coupons"], info["friction_coupons"], info["reserve_coupons"],
+			)
+			for c in range(1, h["coupons_per_wafer"] + 1):
+				coupon_id += 1
+				rel_dir = f"{wafer_dir}/coupon_C{c:02d}"
+				info = reader.dicts(f"{rel_dir}/coupon_info.csv")[0]
+				is_friction = info["run_code"] != "RESERVE"
+				yield "coupons", (
+					coupon_id, info["coupon_code"], wafer_id, batch_ids[b_code],
+					info["grid_row"], info["grid_col"], info["thickness_um"], info["ra_nm"],
+					info["au_wtpct_mean"], run_ids[info["run_code"]] if is_friction else None,
+					info.get("plate"), info.get("probe"), info.get("square"),
+				)
+				for row in reader.dicts(f"{rel_dir}/xrf_map.csv"):
+					yield "xrf_points", (coupon_id, row["grid_x"], row["grid_y"], row["pt_wtpct"], row["au_wtpct"])
+				for row in reader.dicts(f"{rel_dir}/profilometry.csv"):
+					yield "profilometry_points", (coupon_id, row["grid_x"], row["grid_y"], row["thickness_um"])
+				for row in reader.dicts(f"{rel_dir}/nanoindentation.csv"):
+					yield "nanoindentation", (
+						coupon_id, row["indent_id"], row["x_um"], row["y_um"],
+						row["hardness_gpa"], row["reduced_modulus_gpa"], row["max_load_mn"],
+					)
+				afm = reader.dicts(f"{rel_dir}/afm.csv")[0]
+				yield "afm", (coupon_id, afm["ra_nm"], afm["rq_nm"], afm["image_file"])
+				if not is_friction:
+					continue
+				for t in range(1, fa["tracks_per_friction_coupon"] + 1):
+					track_id += 1
+					track_dir = f"{rel_dir}/track_T{t}"
+					tinfo = reader.dicts(f"{track_dir}/track_info.csv")[0]
+					yield "tracks", (
+						track_id, tinfo["track_code"], coupon_id, run_ids[tinfo["run_code"]],
+						tinfo["environment"], tinfo["load_mn"], tinfo["stroke_mm"], tinfo["speed_mm_s"],
+						tinfo["counterface_id"], tinfo["started_at"],
+					)
+					cofs: list[float] = []
+					for row in reader.dicts(f"{track_dir}/cof_vs_cycle.csv"):
+						cofs.append(row["cof"])
+						yield "friction_cycles", (track_id, row["cycle"], row["cof"])
+					ss_mean, ss_std, run_in = summarize_track(cofs)
+					yield "track_summary", (track_id, ss_mean, ss_std, run_in)
+					pt = 0
+					last_cycle = None
+					for row in reader.dicts(f"{track_dir}/friction_loops.csv"):
+						pt = pt + 1 if row["cycle"] == last_cycle else 1
+						last_cycle = row["cycle"]
+						yield "friction_loop_points", (track_id, row["cycle"], pt, row["position_um"], row["friction_force_mn"])
+					wear = reader.dicts(f"{track_dir}/wear.csv")[0]
+					yield "wear", (track_id, wear["wear_volume_um3"], wear["k_archard"], wear["sliding_distance_m"])
+		logger.info("batch %s streamed (through coupon %d, track %d)", b_code, coupon_id, track_id)

common/results.py

@@ -0,0 +1,73 @@
+"""Canonical benchmark result handling (spec 08, test-pipeline-validation).
+
+Every implementation of a benchmark query returns the same canonical rows:
+tuples of str / int / float columns. Rows are compared SORTED; floats agree
+within 1e-9. The canonical expected artifacts are produced from PostgreSQL
+and cross-validated against SQLite (task 08); every measured run (task 09)
+re-validates its rows against them.
+
+Two serializations:
+- full: floats via repr() - lossless, used for stdout and expected/*.rows.csv;
+- rounded (6 decimals): used ONLY for the sha256 fingerprint, so engines
+  whose aggregate arithmetic differs in the last bits hash identically.
+The tolerant row comparison is the authoritative check; the sha256 is a
+compact fingerprint for reports and markers.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import sys
+
+from common.corpus import coerce
+
+FLOAT_TOL = 1e-9
+
+Row = tuple
+
+
+def sort_rows(rows: list[Row]) -> list[Row]:
+	return sorted(rows)
+
+
+def _fmt(value, rounded: bool) -> str:
+	if isinstance(value, float):
+		return f"{round(value, 6):.6f}" if rounded else repr(value)
+	return str(value)
+
+
+def serialize(rows: list[Row], rounded: bool = False) -> str:
+	return "\n".join(",".join(_fmt(v, rounded) for v in row) for row in rows) + "\n"
+
+
+def sha256_of(rows: list[Row]) -> str:
+	return hashlib.sha256(serialize(sort_rows(rows), rounded=True).encode("ascii")).hexdigest()
+
+
+def print_rows(rows: list[Row]) -> None:
+	sys.stdout.write(serialize(sort_rows(rows)))
+
+
+def parse_rows(text: str) -> list[Row]:
+	rows = []
+	for line in text.splitlines():
+		if line:
+			rows.append(tuple(coerce(cell) for cell in line.split(",")))
+	return rows
+
+
+def compare(expected: list[Row], actual: list[Row], tol: float = FLOAT_TOL) -> str | None:
+	"""Return None when equal within tolerance, else a first-difference message."""
+	exp, act = sort_rows(expected), sort_rows(actual)
+	if len(exp) != len(act):
+		return f"row count {len(act)} != expected {len(exp)}"
+	for i, (er, ar) in enumerate(zip(exp, act)):
+		if len(er) != len(ar):
+			return f"row {i}: arity {len(ar)} != {len(er)}"
+		for j, (ev, av) in enumerate(zip(er, ar)):
+			if isinstance(ev, float) or isinstance(av, float):
+				if abs(float(ev) - float(av)) > tol:
+					return f"row {i} col {j}: {av!r} != {ev!r} (tol {tol})"
+			elif ev != av:
+				return f"row {i} col {j}: {av!r} != {ev!r}"
+	return None

common/storage_sizes.py

@@ -0,0 +1,55 @@
+"""Idempotent per-(format, variant) updates to out/bench/storage_sizes.csv.
+
+Contract (docs/rules/build-pipeline-tasks.md section 4): columns
+format,variant,bytes; a task re-run REPLACES its own (format, variant) rows
+and never touches other rows. Replacement is per variant, not per format:
+converters 04-07 own the working-footprint variants while task 09 appends
+archival variants of the same formats.
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import time
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+HEADER = "format,variant,bytes"
+LOCK_TIMEOUT_S = 30.0  # converters 04-07 may run in parallel; the update is a
+LOCK_POLL_S = 0.2      # read-modify-write, so it is serialized via a lock file
+
+
+def update_storage_sizes(bench_dir: Path, fmt: str, variants: list[tuple[str, int]]) -> Path:
+	bench_dir.mkdir(parents=True, exist_ok=True)
+	path = bench_dir / "storage_sizes.csv"
+	lock = bench_dir / "storage_sizes.lock"
+	fd = _acquire(lock)
+	try:
+		replaced = {(fmt, variant) for variant, _ in variants}
+		kept: list[str] = []
+		if path.exists():
+			lines = path.read_text(encoding="ascii").splitlines()
+			if lines and lines[0] != HEADER:
+				raise ValueError(f"{path}: unexpected header {lines[0]!r}")
+			kept = [ln for ln in lines[1:] if ln and tuple(ln.split(",")[:2]) not in replaced]
+		rows = kept + [f"{fmt},{variant},{nbytes}" for variant, nbytes in variants]
+		with open(path, "w", encoding="ascii", newline="\n") as fh:
+			fh.write(HEADER + "\n" + "\n".join(rows) + "\n")
+	finally:
+		os.close(fd)
+		lock.unlink()
+	logger.info("storage_sizes.csv: %s -> %s", fmt, ", ".join(f"{v}={b}" for v, b in variants))
+	return path
+
+
+def _acquire(lock: Path) -> int:
+	deadline = time.monotonic() + LOCK_TIMEOUT_S
+	while True:
+		try:
+			return os.open(lock, os.O_CREAT | os.O_EXCL | os.O_WRONLY)
+		except FileExistsError:
+			if time.monotonic() > deadline:
+				raise TimeoutError(f"storage_sizes lock held too long: {lock}")
+			time.sleep(LOCK_POLL_S)

common/track_summary.py

@@ -0,0 +1,48 @@
+"""Per-track steady-state summary (docs/rules/db-sql-schema.md section 5).
+
+Both engines must produce identical values: SQLite calls this function
+during load (task 05); the PostgreSQL materialized view (task 06)
+implements the same three steps in SQL. Plain sequential float arithmetic
+keeps results within the 1e-9 checksum tolerance across implementations.
+"""
+
+from __future__ import annotations
+
+import math
+
+TAIL_START_CYCLE = 501  # steady-state proxy window: cycles 501..N
+
+
+def summarize_track(cofs: list[float]) -> tuple[float, float, int]:
+	"""Return (cof_ss_mean, cof_ss_std, run_in_cycles) for cofs[i] = cof at cycle i+1."""
+	n = len(cofs)
+	tail = cofs[TAIL_START_CYCLE - 1:]
+	m = _mean(tail)
+	s = _stddev_pop(tail, m)
+
+	run_in = TAIL_START_CYCLE - 1  # fallback when no cycle enters the 2-sigma band
+	band = 2.0 * s
+	for c in range(1, n + 1):
+		if abs(cofs[c - 1] - m) < band:
+			run_in = c
+			break
+
+	ss = cofs[run_in:]  # cycles strictly greater than run_in
+	ss_mean = _mean(ss)
+	ss_std = _stddev_pop(ss, ss_mean)
+	return ss_mean, ss_std, run_in
+
+
+def _mean(values: list[float]) -> float:
+	total = 0.0
+	for v in values:
+		total += v
+	return total / len(values)
+
+
+def _stddev_pop(values: list[float], mean: float) -> float:
+	acc = 0.0
+	for v in values:
+		d = v - mean
+		acc += d * d
+	return math.sqrt(acc / len(values))

convert_json.py

@@ -0,0 +1,293 @@
+"""Task 04 - Convert the CSV corpus to JSON-LD.
+
+Reads ./out/csv/ (validated file-by-file against MANIFEST.csv) and writes
+two ontology-annotated variants:
+- FULL: one compact JSON-LD file per coupon under ./out/json/full/, bulk
+  points embedded. Like spec 07's RDF modeling, bulk points (cycles, loop
+  points, map points) carry no @type - the containing predicate types them;
+  this keeps FULL lean, so its size lands below the 0.7-1.2 GB the spec
+  sketched for a fully typed serialization.
+- HYBRID: single ./out/json/hybrid/dataset.jsonld holding every metadata
+  entity; every per-point array (cycles, loops, xrf, profilometry, indents)
+  is replaced by a {sourceFile, rows, sha256} reference into the CSV tree.
+
+Appends both sizes to ./out/bench/storage_sizes.csv and writes
+./out/.done/04.ok. Spec: docs/specs/04_convert_json.md.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import shutil
+import sys
+from pathlib import Path
+
+from rdflib import Graph
+
+from common.corpus import CorpusReader, ValidationError
+from common.pipeline import (
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	process_metrics,
+	remove_stale_marker,
+	write_marker,
+)
+from common.storage_sizes import update_storage_sizes
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "04"
+DEPENDS_ON = ["01", "03"]
+
+VOCAB = "https://sandia.gov/ontology/tribology#"
+ID_BASE = "https://sandia.gov/ontology/tribology/id/"
+
+CONTEXT = {
+	"@vocab": VOCAB,
+	"tribo": VOCAB,
+	"qudt": "http://qudt.org/schema/qudt/",
+	"unit": "http://qudt.org/vocab/unit/",
+	"prov": "http://www.w3.org/ns/prov#",
+	"batch": ID_BASE + "batch/",
+	"wafer": ID_BASE + "wafer/",
+	"coupon": ID_BASE + "coupon/",
+	"track": ID_BASE + "track/",
+	"run": ID_BASE + "run/",
+	"instrument": ID_BASE + "instrument/",
+	"partOf": {"@id": "tribo:partOf", "@type": "@id"},
+	"cutFrom": {"@id": "tribo:cutFrom", "@type": "@id"},
+	"performedOn": {"@id": "tribo:performedOn", "@type": "@id"},
+	"performedBy": {"@id": "tribo:performedBy", "@type": "@id"},
+	"duringRun": {"@id": "tribo:duringRun", "@type": "@id"},
+	"started_at": {"@id": "prov:startedAtTime"},
+}
+
+MEASUREMENTS = [
+	# (csv file, json key, @type, instrument id, points key)
+	("xrf_map.csv", "xrf", "CompositionMeasurement", "m4_tornado", "points"),
+	("profilometry.csv", "profilometry", "ProfilometryMeasurement", "profilometer", "points"),
+	("nanoindentation.csv", "nanoindentation", "NanoindentationMeasurement", "ti980", "indents"),
+]
+
+
+class JsonConverter:
+	def __init__(self, cfg: dict, csv_root: Path, json_root: Path) -> None:
+		self.cfg = cfg
+		self.json_root = json_root
+		self.reader = CorpusReader(csv_root)
+		self.manifest = self.reader.manifest
+		self.graph: list[dict] = []  # hybrid metadata entities
+		self.full_bytes = 0
+		self.full_files = 0
+		self.counts = {"coupons": 0, "tracks": 0, "cycle_rows": 0, "loop_rows": 0, "xrf_rows": 0}
+
+	def rows_as_dicts(self, relpath: str) -> list[dict]:
+		return self.reader.dicts(relpath)
+
+	def source_ref(self, relpath: str) -> dict:
+		rows, _, sha = self.manifest[relpath]
+		return {"sourceFile": relpath, "rows": rows, "sha256": sha}
+
+	# --- flat entities (hybrid graph) ---
+
+	def convert_flat(self) -> None:
+		for inst in self.cfg["instruments"]:
+			self.graph.append({
+				"@id": f"instrument:{inst['instrument_id']}",
+				"@type": "Instrument",
+				"name": inst["name"],
+				"role": inst["role"],
+			})
+		for row in self.rows_as_dicts("batches.csv"):
+			self.graph.append({"@id": f"batch:{row['batch_code']}", "@type": "Batch", **row})
+			self.graph.append({
+				"@id": f"batch:{row['batch_code']}-simtra",
+				"@type": "SimtraProfile",
+				"performedOn": f"batch:{row['batch_code']}",
+				"performedBy": "instrument:simtra",
+				**self.source_ref(f"simtra/simtra_profile_{row['batch_code']}.csv"),
+			})
+		for row in self.rows_as_dicts("runs.csv"):
+			self.graph.append({"@id": f"run:{row['run_code']}", "@type": "Run", **row})
+
+	# --- per-coupon conversion ---
+
+	def convert_coupons(self) -> None:
+		h = self.cfg["hierarchy"]
+		full_dir = self.json_root / "full"
+		full_dir.mkdir(parents=True, exist_ok=True)
+		for batch in self.cfg["deposition_matrix"]:
+			b_code = batch["batch_code"]
+			for w in range(1, h["wafers_per_batch"] + 1):
+				wafer_code = f"{b_code}-W{w}"
+				wafer_dir = f"batch_{b_code}/wafer_W{w}"
+				info = self.rows_as_dicts(f"{wafer_dir}/wafer_info.csv")[0]
+				self.graph.append({
+					"@id": f"wafer:{wafer_code}", "@type": "Wafer",
+					"partOf": f"batch:{b_code}", **info,
+				})
+				for c in range(1, h["coupons_per_wafer"] + 1):
+					self.convert_coupon(f"{wafer_dir}/coupon_C{c:02d}", full_dir)
+				logger.debug("wafer %s converted", wafer_code)
+			logger.info("batch %s converted: %.1f MiB full so far", b_code, self.full_bytes / (1024 * 1024))
+
+	def convert_coupon(self, rel_dir: str, full_dir: Path) -> None:
+		info = self.rows_as_dicts(f"{rel_dir}/coupon_info.csv")[0]
+		code = info["coupon_code"]
+		coupon_id = f"coupon:{code}"
+		is_friction = info["run_code"] != "RESERVE"
+
+		full_node: dict = {
+			"@context": CONTEXT, "@id": coupon_id, "@type": "Coupon",
+			"cutFrom": f"wafer:{info['wafer_code']}", **info,
+		}
+		hybrid_node: dict = {
+			"@id": coupon_id, "@type": "Coupon",
+			"cutFrom": f"wafer:{info['wafer_code']}", **info,
+		}
+
+		for csv_name, key, cls, inst, points_key in MEASUREMENTS:
+			relpath = f"{rel_dir}/{csv_name}"
+			meta = {"@type": cls, "performedOn": coupon_id, "performedBy": f"instrument:{inst}"}
+			full_node[key] = {**meta, points_key: self.rows_as_dicts(relpath)}
+			hybrid_node[key] = {**meta, **self.source_ref(relpath)}
+		self.counts["xrf_rows"] += self.manifest[f"{rel_dir}/xrf_map.csv"][0]
+
+		afm = self.rows_as_dicts(f"{rel_dir}/afm.csv")[0]
+		afm_node = {"@type": "AFMMeasurement", "performedOn": coupon_id, "performedBy": "instrument:afm", **afm}
+		full_node["afm"] = afm_node
+		hybrid_node["afm"] = afm_node
+
+		if is_friction:
+			tracks_full = []
+			for t in range(1, self.cfg["friction_assignment"]["tracks_per_friction_coupon"] + 1):
+				full_track, hybrid_track = self.convert_track(f"{rel_dir}/track_T{t}", coupon_id, info["run_code"])
+				tracks_full.append(full_track)
+				self.graph.append(hybrid_track)
+			full_node["tracks"] = tracks_full
+
+		out_path = full_dir / f"{code}.jsonld"
+		text = json.dumps(full_node, separators=(",", ":"))
+		data = text.encode("ascii")
+		out_path.write_bytes(data)
+		self.full_bytes += len(data)
+		self.full_files += 1
+		self.graph.append(hybrid_node)
+		self.counts["coupons"] += 1
+
+	def convert_track(self, rel_dir: str, coupon_id: str, run_code: str) -> tuple[dict, dict]:
+		info = self.rows_as_dicts(f"{rel_dir}/track_info.csv")[0]
+		track_id = f"track:{info['track_code']}"
+		base = {
+			"@id": track_id, "@type": "FrictionTest",
+			"partOf": coupon_id, "duringRun": f"run:{run_code}",
+			"performedBy": "instrument:rapid", **info,
+		}
+		wear = self.rows_as_dicts(f"{rel_dir}/wear.csv")[0]
+		wear_node = {"@type": "WearMeasurement", **wear}
+
+		cycles_path = f"{rel_dir}/cof_vs_cycle.csv"
+		loops_path = f"{rel_dir}/friction_loops.csv"
+		full_track = {
+			**base,
+			"cycles": self.rows_as_dicts(cycles_path),
+			"loop_points": self.rows_as_dicts(loops_path),
+			"wear": wear_node,
+		}
+		hybrid_track = {
+			**base,
+			"cycles": self.source_ref(cycles_path),
+			"loop_points": self.source_ref(loops_path),
+			"wear": wear_node,
+		}
+		self.counts["cycle_rows"] += len(full_track["cycles"])
+		self.counts["loop_rows"] += len(full_track["loop_points"])
+		self.counts["tracks"] += 1
+		return full_track, hybrid_track
+
+	# --- outputs ---
+
+	def write_hybrid(self) -> int:
+		hybrid_dir = self.json_root / "hybrid"
+		hybrid_dir.mkdir(parents=True, exist_ok=True)
+		path = hybrid_dir / "dataset.jsonld"
+		with open(path, "w", encoding="ascii", newline="\n") as fh:
+			json.dump({"@context": CONTEXT, "@graph": self.graph}, fh, separators=(",", ":"))
+		return path.stat().st_size
+
+	def validate_counts(self) -> None:
+		v = self.cfg["volumes"]
+		expected = {
+			"coupons": v["coupons_total"],
+			"tracks": v["tracks_total"],
+			"cycle_rows": v["cycle_rows_total"],
+			"loop_rows": v["loop_points_total"],
+			"xrf_rows": v["xrf_points_total"],
+		}
+		for key, exp in expected.items():
+			if self.counts[key] != exp:
+				raise ValidationError(f"count mismatch: {key}: converted {self.counts[key]} != expected {exp}")
+		logger.info("all converted counts match lab_config volumes: %s", self.counts)
+
+
+def validate_sample(paths: list[Path]) -> None:
+	"""Smoke test (spec 04 + test-pipeline-validation): json.load + rdflib parse."""
+	for path in paths:
+		with open(path, encoding="ascii") as fh:
+			json.load(fh)
+		g = Graph()
+		g.parse(path, format="json-ld")
+		if len(g) == 0:
+			raise ValidationError(f"{path}: rdflib parsed 0 triples")
+		logger.info("sample %s: valid JSON, %d triples via rdflib", path.name, len(g))
+
+
+def main() -> int:
+	args = parse_task_args("Task 04: convert the CSV corpus to JSON-LD")
+	out_root: Path = args.out_root
+	json_root = out_root / "json"
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		if json_root.exists():
+			logger.info("re-run: removing previous output %s", json_root)
+			shutil.rmtree(json_root)
+		cfg = load_lab_config(out_root)
+
+		conv = JsonConverter(cfg, out_root / "csv", json_root)
+		conv.convert_flat()
+		conv.convert_coupons()
+		conv.validate_counts()
+		hybrid_bytes = conv.write_hybrid()
+		validate_sample([
+			json_root / "full" / "B722-W2-C13.jsonld",
+			json_root / "hybrid" / "dataset.jsonld",
+		])
+		update_storage_sizes(out_root / "bench", "json", [("full", conv.full_bytes), ("hybrid", hybrid_bytes)])
+		entries = {
+			"full_files": conv.full_files,
+			"full_bytes": conv.full_bytes,
+			"hybrid_bytes": hybrid_bytes,
+			"coupons": conv.counts["coupons"],
+			"tracks": conv.counts["tracks"],
+			"cycle_rows": conv.counts["cycle_rows"],
+			"loop_points": conv.counts["loop_rows"],
+		}
+		entries.update(process_metrics())
+		marker_path = write_marker(out_root, TASK_ID, entries)
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	print(f"task 04 ok: full {conv.full_files} files, {conv.full_bytes / (1024 * 1024):.1f} MiB; "
+		f"hybrid {hybrid_bytes / (1024 * 1024):.1f} MiB")
+	print(f"entities: coupons={conv.counts['coupons']} tracks={conv.counts['tracks']} "
+		f"cycles={conv.counts['cycle_rows']} loop_points={conv.counts['loop_rows']}")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

convert_postgresql.py

@@ -0,0 +1,334 @@
+"""Task 06 - Convert the CSV corpus to PostgreSQL.
+
+Loads the same logical schema as task 05 into the PostgreSQL 16 database
+named by the TRIBO_PG_DSN environment variable: COPY FROM STDIN, RANGE
+partitioning of the two bulk tables by track_id (12 partitions of 120
+tracks), constraints and indexes created after the load (variant B:
+enforced foreign keys everywhere - docs/research/Tribology_FK_Architecture.html),
+track_summary as a materialized view (algorithm fixed in
+docs/rules/db-sql-schema.md section 5, cross-checked against
+common/track_summary.py). Host CPU/RAM during the load is sampled from the
+database host's Prometheus (TRIBO_PROM_URL). Appends live (and, when
+pg_dump is available, dump) sizes to storage_sizes.csv and writes
+./out/.done/06.ok. Spec: docs/specs/06_convert_postgresql.md.
+"""
+
+from __future__ import annotations
+
+import logging
+import shutil
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+import psycopg
+
+from common.corpus import CorpusReader, ValidationError
+from common.monitoring import get_prom_url, host_window_metrics
+from common.pg import get_dsn
+from common.pipeline import (
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	process_metrics,
+	remove_stale_marker,
+	write_marker,
+)
+from common.relational import COLUMNS, TABLES, expected_counts, stream_rows
+from common.storage_sizes import update_storage_sizes
+from common.track_summary import summarize_track
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "06"
+DEPENDS_ON = ["01", "03"]
+
+BATCH_ROWS = 50000  # rows per COPY chunk (mirrors the 50k batching of spec 05)
+PARTITIONS = 12  # RANGE partitions of 120 tracks each, spec 06
+TRACKS_PER_PARTITION = 120
+SUMMARY_PARITY_TRACKS = (1, 720, 1440)  # matview cross-checked vs common/track_summary.py
+PARITY_TOLERANCE = 1e-9  # cross-engine float tolerance, test-pipeline-validation
+
+# Float columns are DOUBLE PRECISION, not REAL: the corpus is float64 and
+# REAL (float4) would break the 1e-9 cross-engine result parity that the
+# whole benchmark depends on (db-sql-schema section 6; spec updated).
+CREATE_TABLES = [
+	"CREATE TABLE instruments (instrument_id INTEGER NOT NULL, instrument_code TEXT NOT NULL, name TEXT NOT NULL, role TEXT NOT NULL)",
+	"CREATE TABLE batches (batch_id INTEGER NOT NULL, batch_code TEXT NOT NULL, pt_gun_tilt_deg INTEGER NOT NULL, au_gun_tilt_deg INTEGER NOT NULL, pt_power_w INTEGER NOT NULL, au_power_w INTEGER NOT NULL, pt_discharge_v INTEGER NOT NULL, au_discharge_v INTEGER NOT NULL, deposition_date DATE NOT NULL)",
+	"CREATE TABLE runs (run_id INTEGER NOT NULL, run_code TEXT NOT NULL, environment TEXT NOT NULL, date DATE NOT NULL, plates INTEGER NOT NULL, operator TEXT NOT NULL)",
+	"CREATE TABLE wafers (wafer_id INTEGER NOT NULL, wafer_code TEXT NOT NULL, batch_id INTEGER NOT NULL, wafer_index INTEGER NOT NULL, deposition_date DATE NOT NULL, coupons INTEGER NOT NULL, friction_coupons INTEGER NOT NULL, reserve_coupons INTEGER NOT NULL)",
+	"CREATE TABLE coupons (coupon_id INTEGER NOT NULL, coupon_code TEXT NOT NULL, wafer_id INTEGER NOT NULL, batch_id INTEGER NOT NULL, grid_row INTEGER NOT NULL, grid_col INTEGER NOT NULL, thickness_um DOUBLE PRECISION NOT NULL, ra_nm DOUBLE PRECISION NOT NULL, au_wtpct_mean DOUBLE PRECISION NOT NULL, run_id INTEGER, plate INTEGER, probe INTEGER, square INTEGER)",
+	"CREATE TABLE tracks (track_id INTEGER NOT NULL, track_code TEXT NOT NULL, coupon_id INTEGER NOT NULL, run_id INTEGER NOT NULL, environment TEXT NOT NULL, load_mn DOUBLE PRECISION NOT NULL, stroke_mm DOUBLE PRECISION NOT NULL, speed_mm_s DOUBLE PRECISION NOT NULL, counterface_id TEXT NOT NULL, started_at TIMESTAMPTZ NOT NULL)",
+	"CREATE TABLE simtra_profiles (batch_id INTEGER NOT NULL, row_no INTEGER NOT NULL, angle_deg DOUBLE PRECISION NOT NULL, energy_ev DOUBLE PRECISION NOT NULL, pt_flux DOUBLE PRECISION NOT NULL, au_flux DOUBLE PRECISION NOT NULL)",
+	"CREATE TABLE xrf_points (coupon_id INTEGER NOT NULL, grid_x INTEGER NOT NULL, grid_y INTEGER NOT NULL, pt_wtpct DOUBLE PRECISION NOT NULL, au_wtpct DOUBLE PRECISION NOT NULL)",
+	"CREATE TABLE profilometry_points (coupon_id INTEGER NOT NULL, grid_x INTEGER NOT NULL, grid_y INTEGER NOT NULL, thickness_um DOUBLE PRECISION NOT NULL)",
+	"CREATE TABLE nanoindentation (coupon_id INTEGER NOT NULL, indent_id INTEGER NOT NULL, x_um DOUBLE PRECISION NOT NULL, y_um DOUBLE PRECISION NOT NULL, hardness_gpa DOUBLE PRECISION NOT NULL, reduced_modulus_gpa DOUBLE PRECISION NOT NULL, max_load_mn DOUBLE PRECISION NOT NULL)",
+	"CREATE TABLE afm (coupon_id INTEGER NOT NULL, ra_nm DOUBLE PRECISION NOT NULL, rq_nm DOUBLE PRECISION NOT NULL, image_file TEXT NOT NULL)",
+	"CREATE TABLE friction_cycles (track_id INTEGER NOT NULL, cycle INTEGER NOT NULL, cof DOUBLE PRECISION NOT NULL) PARTITION BY RANGE (track_id)",
+	"CREATE TABLE friction_loop_points (track_id INTEGER NOT NULL, cycle INTEGER NOT NULL, pt INTEGER NOT NULL, position_um DOUBLE PRECISION NOT NULL, friction_force_mn DOUBLE PRECISION NOT NULL) PARTITION BY RANGE (track_id)",
+	"CREATE TABLE wear (track_id INTEGER NOT NULL, wear_volume_um3 DOUBLE PRECISION NOT NULL, k_archard DOUBLE PRECISION NOT NULL, sliding_distance_m DOUBLE PRECISION NOT NULL)",
+]
+
+CONSTRAINTS = [
+	"ALTER TABLE instruments ADD PRIMARY KEY (instrument_id)",
+	"ALTER TABLE instruments ADD UNIQUE (instrument_code)",
+	"ALTER TABLE batches ADD PRIMARY KEY (batch_id)",
+	"ALTER TABLE batches ADD UNIQUE (batch_code)",
+	"ALTER TABLE runs ADD PRIMARY KEY (run_id)",
+	"ALTER TABLE runs ADD UNIQUE (run_code)",
+	"ALTER TABLE wafers ADD PRIMARY KEY (wafer_id)",
+	"ALTER TABLE wafers ADD UNIQUE (wafer_code)",
+	"ALTER TABLE coupons ADD PRIMARY KEY (coupon_id)",
+	"ALTER TABLE coupons ADD UNIQUE (coupon_code)",
+	"ALTER TABLE tracks ADD PRIMARY KEY (track_id)",
+	"ALTER TABLE tracks ADD UNIQUE (track_code)",
+	"ALTER TABLE simtra_profiles ADD PRIMARY KEY (batch_id, row_no)",
+	"ALTER TABLE xrf_points ADD PRIMARY KEY (coupon_id, grid_x, grid_y)",
+	"ALTER TABLE profilometry_points ADD PRIMARY KEY (coupon_id, grid_x, grid_y)",
+	"ALTER TABLE nanoindentation ADD PRIMARY KEY (coupon_id, indent_id)",
+	"ALTER TABLE afm ADD PRIMARY KEY (coupon_id)",
+	"ALTER TABLE friction_cycles ADD PRIMARY KEY (track_id, cycle)",
+	"ALTER TABLE friction_loop_points ADD PRIMARY KEY (track_id, cycle, pt)",
+	"ALTER TABLE wear ADD PRIMARY KEY (track_id)",
+	"ALTER TABLE wafers ADD FOREIGN KEY (batch_id) REFERENCES batches",
+	"ALTER TABLE coupons ADD FOREIGN KEY (wafer_id) REFERENCES wafers",
+	"ALTER TABLE coupons ADD FOREIGN KEY (batch_id) REFERENCES batches",
+	"ALTER TABLE coupons ADD FOREIGN KEY (run_id) REFERENCES runs",
+	"ALTER TABLE tracks ADD FOREIGN KEY (coupon_id) REFERENCES coupons",
+	"ALTER TABLE tracks ADD FOREIGN KEY (run_id) REFERENCES runs",
+	"ALTER TABLE simtra_profiles ADD FOREIGN KEY (batch_id) REFERENCES batches",
+	"ALTER TABLE xrf_points ADD FOREIGN KEY (coupon_id) REFERENCES coupons",
+	"ALTER TABLE profilometry_points ADD FOREIGN KEY (coupon_id) REFERENCES coupons",
+	"ALTER TABLE nanoindentation ADD FOREIGN KEY (coupon_id) REFERENCES coupons",
+	"ALTER TABLE afm ADD FOREIGN KEY (coupon_id) REFERENCES coupons",
+	"ALTER TABLE friction_cycles ADD FOREIGN KEY (track_id) REFERENCES tracks",
+	"ALTER TABLE friction_loop_points ADD FOREIGN KEY (track_id) REFERENCES tracks",
+	"ALTER TABLE wear ADD FOREIGN KEY (track_id) REFERENCES tracks",
+]
+
+INDEX_DDL = [
+	"CREATE INDEX ix_coupons_au_wtpct_mean ON coupons(au_wtpct_mean)",  # Q2 selective filter
+	"CREATE INDEX ix_tracks_run_id ON tracks(run_id)",  # Q5/Q7 joins to runs
+	"CREATE INDEX ix_tracks_coupon_id ON tracks(coupon_id)",  # Q3/Q6 joins to coupons
+	"CREATE INDEX ix_runs_environment ON runs(environment)",  # Q7 grouping
+	"CREATE INDEX ix_tracks_environment_load ON tracks(environment, load_mn)",  # Q4 filter
+	# Q5 pruning within partitions; cascades to every partition (spec 06).
+	# nanoindentation(coupon_id) from the spec is omitted: the PK prefix serves it.
+	"CREATE INDEX ix_brin_friction_cycles_cycle ON friction_cycles USING BRIN (cycle)",
+]
+
+# Materialized view semantics fixed in docs/rules/db-sql-schema.md section 5.
+MATVIEW_SQL = """
+CREATE MATERIALIZED VIEW track_summary AS
+WITH tail AS (
+	SELECT track_id, avg(cof) AS m, stddev_pop(cof) AS s
+	FROM friction_cycles
+	WHERE cycle >= 501
+	GROUP BY track_id
+),
+run_in AS (
+	SELECT fc.track_id,
+	       COALESCE(MIN(fc.cycle) FILTER (WHERE abs(fc.cof - t.m) < 2 * t.s), 500) AS run_in_cycles
+	FROM friction_cycles fc
+	JOIN tail t USING (track_id)
+	GROUP BY fc.track_id
+)
+SELECT r.track_id,
+       avg(fc.cof) AS cof_ss_mean,
+       stddev_pop(fc.cof) AS cof_ss_std,
+       r.run_in_cycles
+FROM run_in r
+JOIN friction_cycles fc ON fc.track_id = r.track_id AND fc.cycle > r.run_in_cycles
+GROUP BY r.track_id, r.run_in_cycles
+"""
+
+
+def partition_ddl() -> list[str]:
+	ddl = []
+	for parent in ("friction_cycles", "friction_loop_points"):
+		for i in range(PARTITIONS):
+			lo = i * TRACKS_PER_PARTITION + 1
+			hi = lo + TRACKS_PER_PARTITION
+			ddl.append(f"CREATE TABLE {parent}_p{i + 1:02d} PARTITION OF {parent} FOR VALUES FROM ({lo}) TO ({hi})")
+	return ddl
+
+
+class PgLoader:
+	def __init__(self, conn: psycopg.Connection) -> None:
+		self.conn = conn
+		self.buffers: dict[str, list[tuple]] = {t: [] for t in TABLES}
+		self.summary_parity: dict[int, tuple] = {}
+
+	def create_schema(self) -> None:
+		with self.conn.cursor() as cur:
+			cur.execute("DROP MATERIALIZED VIEW IF EXISTS track_summary")
+			for table in reversed(TABLES):
+				if table != "track_summary":
+					cur.execute(f"DROP TABLE IF EXISTS {table} CASCADE")
+			for ddl in CREATE_TABLES + partition_ddl():
+				cur.execute(ddl)
+		self.conn.commit()
+		logger.info("schema created: %d tables, %d partitions", len(CREATE_TABLES), 2 * PARTITIONS)
+
+	def load(self, cfg: dict, reader: CorpusReader) -> None:
+		pending = 0
+		for table, row in stream_rows(cfg, reader):
+			if table == "track_summary":
+				# computed server-side as a materialized view; keep a sample
+				# of the Python values for the cross-engine parity check
+				if row[0] in SUMMARY_PARITY_TRACKS:
+					self.summary_parity[row[0]] = row
+				continue
+			self.buffers[table].append(row)
+			pending += 1
+			if pending >= BATCH_ROWS:
+				self.flush_all()
+				pending = 0
+		self.flush_all()
+
+	def flush_all(self) -> None:
+		for table in TABLES:
+			buf = self.buffers.get(table)
+			if not buf:
+				continue
+			cols = ", ".join(COLUMNS[table])
+			with self.conn.cursor() as cur:
+				with cur.copy(f"COPY {table} ({cols}) FROM STDIN") as copy:
+					for row in buf:
+						copy.write_row(row)
+			buf.clear()
+		self.conn.commit()
+
+	def finalize(self) -> None:
+		with self.conn.cursor() as cur:
+			for ddl in CONSTRAINTS:
+				cur.execute(ddl)
+			for ddl in INDEX_DDL:
+				cur.execute(ddl)
+			cur.execute(MATVIEW_SQL)
+			cur.execute("CREATE UNIQUE INDEX ix_track_summary_track_id ON track_summary(track_id)")  # Q2/Q3 joins
+		self.conn.commit()
+		logger.info("constraints, indexes and track_summary materialized view created")
+
+	def validate(self, cfg: dict) -> dict[str, int]:
+		counts: dict[str, int] = {}
+		with self.conn.cursor() as cur:
+			for table, exp in expected_counts(cfg).items():
+				got = cur.execute(f"SELECT COUNT(*) FROM {table}").fetchone()[0]
+				counts[table] = got
+				if got != exp:
+					raise ValidationError(f"row count mismatch: {table}: loaded {got} != expected {exp}")
+			reserve = cur.execute("SELECT COUNT(*) FROM coupons WHERE run_id IS NULL").fetchone()[0]
+			if reserve != cfg["friction_assignment"]["reserve_coupons_total"]:
+				raise ValidationError(f"reserve coupons: {reserve} != expected")
+			for track_id, py_row in self.summary_parity.items():
+				db_row = cur.execute(
+					"SELECT cof_ss_mean, cof_ss_std, run_in_cycles FROM track_summary WHERE track_id = %s",
+					(track_id,),
+				).fetchone()
+				if db_row is None:
+					raise ValidationError(f"track_summary missing track {track_id}")
+				if (abs(db_row[0] - py_row[1]) > PARITY_TOLERANCE
+						or abs(db_row[1] - py_row[2]) > PARITY_TOLERANCE
+						or db_row[2] != py_row[3]):
+					raise ValidationError(
+						f"track_summary parity failed for track {track_id}: db={db_row} py={py_row[1:]}"
+					)
+		logger.info(
+			"all %d table counts match, %d reserve coupons, matview parity ok for tracks %s",
+			len(counts), reserve, list(self.summary_parity),
+		)
+		return counts
+
+	def table_sizes(self) -> dict[str, int]:
+		sizes: dict[str, int] = {}
+		with self.conn.cursor() as cur:
+			for table in TABLES:
+				total = cur.execute(
+					"""SELECT COALESCE(pg_total_relation_size(%s::regclass), 0)
+					 + COALESCE((SELECT sum(pg_total_relation_size(inhrelid))
+					             FROM pg_inherits WHERE inhparent = %s::regclass), 0)""",
+					(table, table),
+				).fetchone()[0]
+				sizes[table] = int(total)
+		return sizes
+
+
+def make_dump(dsn: str, dump_path: Path) -> int:
+	pg_dump = shutil.which("pg_dump")
+	if not pg_dump:
+		logger.warning("pg_dump skipped (client not installed on this host; produce the dump on the database host)")
+		return 0
+	dump_path.parent.mkdir(parents=True, exist_ok=True)
+	subprocess.run([pg_dump, "--format=custom", f"--file={dump_path}", dsn], check=True)
+	return dump_path.stat().st_size
+
+
+def main() -> int:
+	args = parse_task_args("Task 06: convert the CSV corpus to PostgreSQL")
+	out_root: Path = args.out_root
+	dsn = get_dsn()
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		cfg = load_lab_config(out_root)
+
+		t_start = time.time()
+		with psycopg.connect(dsn) as conn:
+			database = conn.info.dbname
+			logger.info("connected to database %r on %s", database, conn.info.host)
+			loader = PgLoader(conn)
+			loader.create_schema()
+			loader.load(cfg, CorpusReader(out_root / "csv"))
+			load_done = time.time()
+			loader.finalize()
+			counts = loader.validate(cfg)
+			sizes = loader.table_sizes()
+		with psycopg.connect(dsn, autocommit=True) as conn:
+			conn.execute("VACUUM ANALYZE")
+		t_end = time.time()
+
+		live_bytes = sum(sizes.values())
+		variants = [("live", live_bytes)]
+		dump_bytes = make_dump(dsn, out_root / "pg" / "tribo.dump")
+		if dump_bytes:
+			variants.append(("dump", dump_bytes))
+		update_storage_sizes(out_root / "bench", "pg", variants)
+
+		entries = {
+			"database": database,
+			"tables": len(counts),
+			"total_rows": sum(counts.values()),
+			"live_bytes": live_bytes,
+			"friction_cycles_bytes": sizes["friction_cycles"],
+			"friction_loop_points_bytes": sizes["friction_loop_points"],
+			"copy_seconds": round(load_done - t_start, 1),
+			"total_seconds": round(t_end - t_start, 1),
+			"dump": dump_bytes if dump_bytes else "skipped (pg_dump unavailable on this host)",
+		}
+		entries.update(process_metrics())  # client-side converter cost
+		prom_url = get_prom_url()
+		if prom_url:
+			metrics = host_window_metrics(prom_url, t_start, t_end)
+			if metrics:
+				entries.update(metrics)
+				logger.info("db host during load: %s", metrics)
+		marker_path = write_marker(out_root, TASK_ID, entries)
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	print(f"task 06 ok: database {database} ({live_bytes / (1024 * 1024):.1f} MiB live, "
+		f"{len(counts)} tables, {sum(counts.values())} rows, {entries['total_seconds']} s)")
+	print(f"bulk rows: cycles={counts['friction_cycles']} loop_points={counts['friction_loop_points']} "
+		f"track_summary={counts['track_summary']}")
+	if prom_url and "host_cpu_avg_pct" in entries:
+		print(f"db host: cpu avg {entries['host_cpu_avg_pct']}% max {entries['host_cpu_max_pct']}%, "
+			f"ram peak {entries['host_ram_used_peak_mb']} MB (delta {entries['host_ram_used_delta_mb']} MB)")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

convert_rdf.py

@@ -0,0 +1,395 @@
+"""Task 07 - Convert the CSV corpus to RDF.
+
+Streams the corpus into ./out/rdf/dataset.nt.gz (N-Triples, gzip) and
+./out/rdf/dataset.ttl (Turtle) with the task 04 vocabulary (tribo# + PROV-O
+for started_at), then bulk-loads the N-Triples into an embedded on-disk
+oxigraph store (./out/rdf/oxigraph_store/) and smoke-tests it with SPARQL
+COUNT queries. Bulk points (cycles, loop points, map points) carry no
+rdf:type - the containing predicate types them (documented decision per
+spec 07), which is why the triple count lands below the spec's 25-45M
+sketch. Appends serialization and store sizes to storage_sizes.csv and
+writes ./out/.done/07.ok. Spec: docs/specs/07_convert_rdf.md.
+"""
+
+from __future__ import annotations
+
+import gzip
+import io
+import logging
+import shutil
+import sys
+import time
+from pathlib import Path
+
+from pyoxigraph import Store
+
+try:
+	from pyoxigraph import RdfFormat
+	NT_FORMAT = RdfFormat.N_TRIPLES
+except ImportError:  # pyoxigraph < 0.5 uses MIME strings
+	NT_FORMAT = "application/n-triples"
+
+from common.corpus import CorpusReader, ValidationError
+from common.pipeline import (
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	process_metrics,
+	remove_stale_marker,
+	write_marker,
+)
+from common.storage_sizes import update_storage_sizes
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "07"
+DEPENDS_ON = ["01", "03"]
+
+VOCAB = "https://sandia.gov/ontology/tribology#"
+ID_BASE = "https://sandia.gov/ontology/tribology/id/"
+XSD = "http://www.w3.org/2001/XMLSchema#"
+
+PREFIXES = {
+	"tribo": VOCAB,
+	"batch": ID_BASE + "batch/",
+	"wafer": ID_BASE + "wafer/",
+	"coupon": ID_BASE + "coupon/",
+	"track": ID_BASE + "track/",
+	"run": ID_BASE + "run/",
+	"instrument": ID_BASE + "instrument/",
+	"simtra": ID_BASE + "simtra/",
+	"prov": "http://www.w3.org/ns/prov#",
+	"xsd": XSD,
+}
+
+INT_COLS = {
+	"cycle", "grid_x", "grid_y", "indent_id", "plates", "plate", "probe", "square",
+	"row_no", "wafer_index", "coupons", "friction_coupons", "reserve_coupons",
+	"pt_gun_tilt_deg", "au_gun_tilt_deg", "pt_power_w", "au_power_w",
+	"pt_discharge_v", "au_discharge_v",
+}
+DOUBLE_COLS = {
+	"cof", "position_um", "friction_force_mn", "pt_wtpct", "au_wtpct", "thickness_um",
+	"x_um", "y_um", "hardness_gpa", "reduced_modulus_gpa", "max_load_mn", "ra_nm",
+	"rq_nm", "wear_volume_um3", "k_archard", "sliding_distance_m", "load_mn",
+	"stroke_mm", "speed_mm_s", "angle_deg", "energy_ev", "pt_flux", "au_flux",
+	"au_wtpct_mean",
+}
+DATE_COLS = {"date", "deposition_date"}
+
+MEASUREMENTS = [
+	("xrf_map.csv", "xrf", "CompositionMeasurement", "m4_tornado"),
+	("profilometry.csv", "profilometry", "ProfilometryMeasurement", "profilometer"),
+	("nanoindentation.csv", "nanoindentation", "NanoindentationMeasurement", "ti980"),
+]
+
+
+def lit(col: str, token: str) -> tuple:
+	if col in INT_COLS:
+		return ("l", token, "integer")
+	if col in DOUBLE_COLS:
+		return ("l", token, "double")
+	if col in DATE_COLS:
+		return ("l", token, "date")
+	if col == "started_at":
+		return ("l", token, "dateTime")
+	return ("l", token, None)
+
+
+def iri(curie: str) -> tuple:
+	return ("i", curie)
+
+
+def node(pairs: list) -> tuple:
+	return ("n", pairs)
+
+
+class RdfEmitter:
+	"""Writes the same subject blocks to N-Triples and Turtle streams."""
+
+	def __init__(self, nt_fh, ttl_fh) -> None:
+		self.nt = nt_fh
+		self.ttl = ttl_fh
+		self.triples = 0
+		self._blank = 0
+		for prefix, base in PREFIXES.items():
+			self.ttl.write(f"@prefix {prefix}: <{base}> .\n")
+		self.ttl.write("\n")
+
+	def _full(self, curie: str) -> str:
+		prefix, local = curie.split(":", 1)
+		return f"<{PREFIXES[prefix]}{local}>"
+
+	def _pred_nt(self, pred: str) -> str:
+		if pred == "a":
+			return "<http://www.w3.org/1999/02/22-rdf-syntax-ns#type>"
+		return self._full(pred)
+
+	@staticmethod
+	def _escape(text: str) -> str:
+		return text.replace("\\", "\\\\").replace('"', '\\"')
+
+	def _obj_nt(self, obj: tuple) -> str:
+		kind = obj[0]
+		if kind == "i":
+			return self._full(obj[1])
+		if kind == "l":
+			_, token, dtype = obj
+			if dtype is None:
+				return f'"{self._escape(token)}"'
+			return f'"{token}"^^<{XSD}{dtype}>'
+		self._blank += 1
+		label = f"_:b{self._blank}"
+		for pred, sub_obj in obj[1]:
+			self.nt.write(f"{label} {self._pred_nt(pred)} {self._obj_nt(sub_obj)} .\n")
+			self.triples += 1
+		return label
+
+	def _obj_ttl(self, obj: tuple) -> str:
+		kind = obj[0]
+		if kind == "i":
+			return obj[1]
+		if kind == "l":
+			_, token, dtype = obj
+			if dtype is None:
+				return f'"{self._escape(token)}"'
+			if dtype == "integer":
+				return token
+			return f'"{token}"^^xsd:{dtype}'
+		inner = " ; ".join(f"{p} {self._obj_ttl(o)}" for p, o in obj[1])
+		return f"[ {inner} ]"
+
+	def emit(self, subject: str, pairs: list) -> None:
+		subj_nt = self._full(subject)
+		ttl_parts = []
+		for pred, obj in pairs:
+			# N-Triples: nested nodes expand to labelled blanks first, so the
+			# referencing triple is written after the node's own triples.
+			obj_nt = self._obj_nt(obj)
+			self.nt.write(f"{subj_nt} {self._pred_nt(pred)} {obj_nt} .\n")
+			self.triples += 1
+			ttl_parts.append(f"{pred} {self._obj_ttl(obj)}")
+		self.ttl.write(subject + " " + " ;\n  ".join(ttl_parts) + " .\n")
+
+
+class RdfConverter:
+	def __init__(self, cfg: dict, csv_root: Path, emitter: RdfEmitter) -> None:
+		self.cfg = cfg
+		self.reader = CorpusReader(csv_root)
+		self.em = emitter
+		self.counts = {"coupons": 0, "tracks": 0, "cycle_nodes": 0, "loop_nodes": 0, "xrf_nodes": 0}
+
+	def raw_rows(self, relpath: str) -> list[dict[str, str]]:
+		header, rows = self.reader.read(relpath)
+		keys = [h.lower() for h in header]
+		return [dict(zip(keys, row)) for row in rows]
+
+	def prop_pairs(self, row: dict[str, str], skip: tuple = ()) -> list:
+		return [
+			("prov:startedAtTime" if col == "started_at" else f"tribo:{col}", lit(col, token))
+			for col, token in row.items()
+			if col not in skip and token != ""
+		]
+
+	def convert(self) -> None:
+		for inst in self.cfg["instruments"]:
+			self.em.emit(f"instrument:{inst['instrument_id']}", [
+				("a", iri("tribo:Instrument")),
+				("tribo:name", ("l", inst["name"], None)),
+				("tribo:role", ("l", inst["role"], None)),
+			])
+		for row in self.raw_rows("batches.csv"):
+			code = row["batch_code"]
+			self.em.emit(f"batch:{code}", [("a", iri("tribo:Batch"))] + self.prop_pairs(row))
+			points = [
+				("tribo:hasPoint", node([("tribo:row_no", ("l", str(no), "integer"))] + self.prop_pairs(p)))
+				for no, p in enumerate(self.raw_rows(f"simtra/simtra_profile_{code}.csv"), 1)
+			]
+			self.em.emit(f"simtra:{code}", [
+				("a", iri("tribo:SimtraProfile")),
+				("tribo:performedOn", iri(f"batch:{code}")),
+				("tribo:performedBy", iri("instrument:simtra")),
+			] + points)
+		for row in self.raw_rows("runs.csv"):
+			self.em.emit(f"run:{row['run_code']}", [("a", iri("tribo:Run"))] + self.prop_pairs(row))
+
+		h = self.cfg["hierarchy"]
+		for batch in self.cfg["deposition_matrix"]:
+			b_code = batch["batch_code"]
+			for w in range(1, h["wafers_per_batch"] + 1):
+				wafer_dir = f"batch_{b_code}/wafer_W{w}"
+				info = self.raw_rows(f"{wafer_dir}/wafer_info.csv")[0]
+				self.em.emit(f"wafer:{info['wafer_code']}", [
+					("a", iri("tribo:Wafer")),
+					("tribo:partOf", iri(f"batch:{b_code}")),
+				] + self.prop_pairs(info, skip=("batch_code",)))
+				for c in range(1, h["coupons_per_wafer"] + 1):
+					self.convert_coupon(f"{wafer_dir}/coupon_C{c:02d}")
+			logger.info("batch %s converted (%d triples so far)", b_code, self.em.triples)
+
+	def convert_coupon(self, rel_dir: str) -> None:
+		info = self.raw_rows(f"{rel_dir}/coupon_info.csv")[0]
+		code = info["coupon_code"]
+		subject = f"coupon:{code}"
+		is_friction = info["run_code"] != "RESERVE"
+		pairs = [("a", iri("tribo:Coupon")), ("tribo:cutFrom", iri(f"wafer:{info['wafer_code']}"))]
+		pairs += self.prop_pairs(info, skip=("batch_code", "wafer_code", "run_code"))
+		pairs.append(("tribo:run_code", ("l", info["run_code"], None)))
+		if is_friction:
+			pairs.append(("tribo:duringRun", iri(f"run:{info['run_code']}")))
+
+		for csv_name, key, cls, inst in MEASUREMENTS:
+			points = [
+				("tribo:hasPoint" if key != "nanoindentation" else "tribo:hasIndent", node(self.prop_pairs(p)))
+				for p in self.raw_rows(f"{rel_dir}/{csv_name}")
+			]
+			if key == "xrf":
+				self.counts["xrf_nodes"] += len(points)
+			pairs.append((f"tribo:{key}", node([
+				("a", iri(f"tribo:{cls}")),
+				("tribo:performedOn", iri(subject)),
+				("tribo:performedBy", iri(f"instrument:{inst}")),
+			] + points)))
+		afm = self.raw_rows(f"{rel_dir}/afm.csv")[0]
+		pairs.append(("tribo:afm", node([
+			("a", iri("tribo:AFMMeasurement")),
+			("tribo:performedOn", iri(subject)),
+			("tribo:performedBy", iri("instrument:afm")),
+		] + self.prop_pairs(afm))))
+		self.em.emit(subject, pairs)
+		self.counts["coupons"] += 1
+
+		if is_friction:
+			for t in range(1, self.cfg["friction_assignment"]["tracks_per_friction_coupon"] + 1):
+				self.convert_track(f"{rel_dir}/track_T{t}", subject)
+
+	def convert_track(self, rel_dir: str, coupon_subject: str) -> None:
+		info = self.raw_rows(f"{rel_dir}/track_info.csv")[0]
+		pairs = [
+			("a", iri("tribo:FrictionTest")),
+			("tribo:partOf", iri(coupon_subject)),
+			("tribo:duringRun", iri(f"run:{info['run_code']}")),
+			("tribo:performedBy", iri("instrument:rapid")),
+		] + self.prop_pairs(info, skip=("track_code", "run_code"))
+		pairs.insert(1, ("tribo:track_code", ("l", info["track_code"], None)))
+
+		cycles = self.raw_rows(f"{rel_dir}/cof_vs_cycle.csv")
+		pairs += [("tribo:hasCycle", node(self.prop_pairs(row))) for row in cycles]
+		self.counts["cycle_nodes"] += len(cycles)
+		loops = self.raw_rows(f"{rel_dir}/friction_loops.csv")
+		pairs += [("tribo:hasLoopPoint", node(self.prop_pairs(row))) for row in loops]
+		self.counts["loop_nodes"] += len(loops)
+		wear = self.raw_rows(f"{rel_dir}/wear.csv")[0]
+		pairs.append(("tribo:wear", node([("a", iri("tribo:WearMeasurement"))] + self.prop_pairs(wear))))
+		self.em.emit(f"track:{info['track_code']}", pairs)
+		self.counts["tracks"] += 1
+
+	def validate_counts(self) -> None:
+		v = self.cfg["volumes"]
+		expected = {
+			"coupons": v["coupons_total"],
+			"tracks": v["tracks_total"],
+			"cycle_nodes": v["cycle_rows_total"],
+			"loop_nodes": v["loop_points_total"],
+			"xrf_nodes": v["xrf_points_total"],
+		}
+		for key, exp in expected.items():
+			if self.counts[key] != exp:
+				raise ValidationError(f"count mismatch: {key}: converted {self.counts[key]} != expected {exp}")
+		logger.info("all converted counts match lab_config volumes: %s", self.counts)
+
+
+def bulk_load(store: Store, nt_gz: Path) -> None:
+	with gzip.open(nt_gz, "rb") as fh:
+		try:
+			store.bulk_load(fh, NT_FORMAT)
+		except TypeError:
+			store.bulk_load(input=fh, format=NT_FORMAT)
+
+
+def sparql_count(store: Store, query: str) -> int:
+	solutions = store.query(query)
+	return int(next(iter(solutions))[0].value)
+
+
+def main() -> int:
+	args = parse_task_args("Task 07: convert the CSV corpus to RDF")
+	out_root: Path = args.out_root
+	rdf_root = out_root / "rdf"
+	nt_gz = rdf_root / "dataset.nt.gz"
+	ttl_path = rdf_root / "dataset.ttl"
+	store_dir = rdf_root / "oxigraph_store"
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		if rdf_root.exists():
+			logger.info("re-run: removing previous output %s", rdf_root)
+			shutil.rmtree(rdf_root)
+		rdf_root.mkdir(parents=True)
+		cfg = load_lab_config(out_root)
+
+		# mtime=0 keeps the gzip byte-identical across re-runs (data-determinism)
+		with open(nt_gz, "wb") as raw:
+			gz = gzip.GzipFile(filename="", mode="wb", fileobj=raw, compresslevel=6, mtime=0)
+			with io.TextIOWrapper(gz, encoding="ascii", newline="\n") as nt_fh, \
+					open(ttl_path, "w", encoding="ascii", newline="\n") as ttl_fh:
+				emitter = RdfEmitter(nt_fh, ttl_fh)
+				conv = RdfConverter(cfg, out_root / "csv", emitter)
+				conv.convert()
+				conv.validate_counts()
+				triples = emitter.triples
+		nt_bytes = nt_gz.stat().st_size
+		ttl_bytes = ttl_path.stat().st_size
+		logger.info("serialized %d triples: nt.gz %.1f MiB, ttl %.1f MiB",
+			triples, nt_bytes / 1048576, ttl_bytes / 1048576)
+
+		t0 = time.time()
+		store = Store(str(store_dir))
+		bulk_load(store, nt_gz)
+		store.flush()
+		load_seconds = round(time.time() - t0, 1)
+
+		total = sparql_count(store, "SELECT (COUNT(*) AS ?n) WHERE { ?s ?p ?o }")
+		if total != triples:
+			raise ValidationError(f"store holds {total} triples, serialized {triples}")
+		q1 = sparql_count(
+			store,
+			f"SELECT (COUNT(?c) AS ?n) WHERE {{ <{ID_BASE}track/B722-W2-C13-T2> <{VOCAB}hasCycle> ?c }}",
+		)
+		if q1 != cfg["volumes"]["cycles_per_track"]:
+			raise ValidationError(f"SPARQL Q1 smoke test returned {q1} cycles")
+		logger.info("store smoke tests ok: %d triples, Q1 track has %d cycles (load %.1f s)", total, q1, load_seconds)
+		del store
+		store_bytes = sum(f.stat().st_size for f in store_dir.rglob("*") if f.is_file())
+
+		update_storage_sizes(out_root / "bench", "rdf", [
+			("nt_gz", nt_bytes), ("ttl", ttl_bytes), ("store", store_bytes),
+		])
+		entries = {
+			"triples": triples,
+			"nt_gz_bytes": nt_bytes,
+			"ttl_bytes": ttl_bytes,
+			"store_bytes": store_bytes,
+			"store_load_seconds": load_seconds,
+			"coupons": conv.counts["coupons"],
+			"tracks": conv.counts["tracks"],
+			"cycle_nodes": conv.counts["cycle_nodes"],
+			"loop_nodes": conv.counts["loop_nodes"],
+		}
+		entries.update(process_metrics())
+		marker_path = write_marker(out_root, TASK_ID, entries)
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	print(f"task 07 ok: {triples} triples; nt.gz {nt_bytes / 1048576:.1f} MiB, "
+		f"ttl {ttl_bytes / 1048576:.1f} MiB, store {store_bytes / 1048576:.1f} MiB "
+		f"(bulk load {load_seconds} s)")
+	print(f"entities: coupons={conv.counts['coupons']} tracks={conv.counts['tracks']} "
+		f"cycles={conv.counts['cycle_nodes']} loop_points={conv.counts['loop_nodes']}")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

convert_sqlite.py

@@ -0,0 +1,286 @@
+"""Task 05 - Convert the CSV corpus to SQLite.
+
+Builds ./out/sqlite/tribo.db: variant B key schema (composite natural PKs
+on bulk tables, WITHOUT ROWID, enforced foreign keys - see
+docs/research/Tribology_FK_Architecture.html), Q1-Q7 indexes created after
+the load, track_summary precomputed. Every consumed CSV is verified
+against MANIFEST.csv. Appends the db size to storage_sizes.csv and writes
+./out/.done/05.ok. Spec: docs/specs/05_convert_sqlite.md.
+"""
+
+from __future__ import annotations
+
+import logging
+import sqlite3
+import sys
+from pathlib import Path
+
+from common.corpus import CorpusReader, ValidationError
+from common.pipeline import (
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	process_metrics,
+	remove_stale_marker,
+	write_marker,
+)
+from common.relational import COLUMNS, TABLES, expected_counts, stream_rows
+from common.storage_sizes import update_storage_sizes
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "05"
+DEPENDS_ON = ["01", "03"]
+
+BATCH_ROWS = 50000  # rows per executemany transaction, spec 05
+LOAD_PRAGMAS = [
+	"PRAGMA journal_mode=OFF",
+	"PRAGMA synchronous=OFF",
+	"PRAGMA cache_size=-262144",
+	"PRAGMA foreign_keys=ON",
+]
+
+DDL = [
+	"""CREATE TABLE instruments (
+		instrument_id INTEGER PRIMARY KEY,
+		instrument_code TEXT NOT NULL UNIQUE,
+		name TEXT NOT NULL,
+		role TEXT NOT NULL
+	)""",
+	"""CREATE TABLE batches (
+		batch_id INTEGER PRIMARY KEY,
+		batch_code TEXT NOT NULL UNIQUE,
+		pt_gun_tilt_deg INTEGER NOT NULL,
+		au_gun_tilt_deg INTEGER NOT NULL,
+		pt_power_w INTEGER NOT NULL,
+		au_power_w INTEGER NOT NULL,
+		pt_discharge_v INTEGER NOT NULL,
+		au_discharge_v INTEGER NOT NULL,
+		deposition_date TEXT NOT NULL
+	)""",
+	"""CREATE TABLE runs (
+		run_id INTEGER PRIMARY KEY,
+		run_code TEXT NOT NULL UNIQUE,
+		environment TEXT NOT NULL,
+		date TEXT NOT NULL,
+		plates INTEGER NOT NULL,
+		operator TEXT NOT NULL
+	)""",
+	"""CREATE TABLE wafers (
+		wafer_id INTEGER PRIMARY KEY,
+		wafer_code TEXT NOT NULL UNIQUE,
+		batch_id INTEGER NOT NULL REFERENCES batches(batch_id),
+		wafer_index INTEGER NOT NULL,
+		deposition_date TEXT NOT NULL,
+		coupons INTEGER NOT NULL,
+		friction_coupons INTEGER NOT NULL,
+		reserve_coupons INTEGER NOT NULL
+	)""",
+	"""CREATE TABLE coupons (
+		coupon_id INTEGER PRIMARY KEY,
+		coupon_code TEXT NOT NULL UNIQUE,
+		wafer_id INTEGER NOT NULL REFERENCES wafers(wafer_id),
+		batch_id INTEGER NOT NULL REFERENCES batches(batch_id),
+		grid_row INTEGER NOT NULL,
+		grid_col INTEGER NOT NULL,
+		thickness_um REAL NOT NULL,
+		ra_nm REAL NOT NULL,
+		au_wtpct_mean REAL NOT NULL,
+		run_id INTEGER REFERENCES runs(run_id),
+		plate INTEGER,
+		probe INTEGER,
+		square INTEGER
+	)""",
+	"""CREATE TABLE tracks (
+		track_id INTEGER PRIMARY KEY,
+		track_code TEXT NOT NULL UNIQUE,
+		coupon_id INTEGER NOT NULL REFERENCES coupons(coupon_id),
+		run_id INTEGER NOT NULL REFERENCES runs(run_id),
+		environment TEXT NOT NULL,
+		load_mn REAL NOT NULL,
+		stroke_mm REAL NOT NULL,
+		speed_mm_s REAL NOT NULL,
+		counterface_id TEXT NOT NULL,
+		started_at TEXT NOT NULL
+	)""",
+	"""CREATE TABLE simtra_profiles (
+		batch_id INTEGER NOT NULL REFERENCES batches(batch_id),
+		row_no INTEGER NOT NULL,
+		angle_deg REAL NOT NULL,
+		energy_ev REAL NOT NULL,
+		pt_flux REAL NOT NULL,
+		au_flux REAL NOT NULL,
+		PRIMARY KEY (batch_id, row_no)
+	) WITHOUT ROWID""",
+	"""CREATE TABLE xrf_points (
+		coupon_id INTEGER NOT NULL REFERENCES coupons(coupon_id),
+		grid_x INTEGER NOT NULL,
+		grid_y INTEGER NOT NULL,
+		pt_wtpct REAL NOT NULL,
+		au_wtpct REAL NOT NULL,
+		PRIMARY KEY (coupon_id, grid_x, grid_y)
+	) WITHOUT ROWID""",
+	"""CREATE TABLE profilometry_points (
+		coupon_id INTEGER NOT NULL REFERENCES coupons(coupon_id),
+		grid_x INTEGER NOT NULL,
+		grid_y INTEGER NOT NULL,
+		thickness_um REAL NOT NULL,
+		PRIMARY KEY (coupon_id, grid_x, grid_y)
+	) WITHOUT ROWID""",
+	"""CREATE TABLE nanoindentation (
+		coupon_id INTEGER NOT NULL REFERENCES coupons(coupon_id),
+		indent_id INTEGER NOT NULL,
+		x_um REAL NOT NULL,
+		y_um REAL NOT NULL,
+		hardness_gpa REAL NOT NULL,
+		reduced_modulus_gpa REAL NOT NULL,
+		max_load_mn REAL NOT NULL,
+		PRIMARY KEY (coupon_id, indent_id)
+	) WITHOUT ROWID""",
+	"""CREATE TABLE afm (
+		coupon_id INTEGER PRIMARY KEY REFERENCES coupons(coupon_id),
+		ra_nm REAL NOT NULL,
+		rq_nm REAL NOT NULL,
+		image_file TEXT NOT NULL
+	)""",
+	"""CREATE TABLE friction_cycles (
+		track_id INTEGER NOT NULL REFERENCES tracks(track_id),
+		cycle INTEGER NOT NULL,
+		cof REAL NOT NULL,
+		PRIMARY KEY (track_id, cycle)
+	) WITHOUT ROWID""",
+	"""CREATE TABLE friction_loop_points (
+		track_id INTEGER NOT NULL REFERENCES tracks(track_id),
+		cycle INTEGER NOT NULL,
+		pt INTEGER NOT NULL,
+		position_um REAL NOT NULL,
+		friction_force_mn REAL NOT NULL,
+		PRIMARY KEY (track_id, cycle, pt)
+	) WITHOUT ROWID""",
+	"""CREATE TABLE wear (
+		track_id INTEGER PRIMARY KEY REFERENCES tracks(track_id),
+		wear_volume_um3 REAL NOT NULL,
+		k_archard REAL NOT NULL,
+		sliding_distance_m REAL NOT NULL
+	)""",
+	"""CREATE TABLE track_summary (
+		track_id INTEGER PRIMARY KEY REFERENCES tracks(track_id),
+		cof_ss_mean REAL NOT NULL,
+		cof_ss_std REAL NOT NULL,
+		run_in_cycles INTEGER NOT NULL
+	)""",
+]
+
+INDEX_DDL = [
+	"CREATE INDEX ix_coupons_au_wtpct_mean ON coupons(au_wtpct_mean)",  # Q2 selective filter
+	"CREATE INDEX ix_tracks_run_id ON tracks(run_id)",  # Q5/Q7 joins to runs
+	"CREATE INDEX ix_tracks_coupon_id ON tracks(coupon_id)",  # Q3/Q6 joins to coupons
+	"CREATE INDEX ix_runs_environment ON runs(environment)",  # Q7 grouping
+	"CREATE INDEX ix_tracks_environment_load ON tracks(environment, load_mn)",  # Q4 filter
+]
+
+
+class SqliteLoader:
+	def __init__(self, db_path: Path) -> None:
+		db_path.parent.mkdir(parents=True, exist_ok=True)
+		self.conn = sqlite3.connect(db_path)
+		for pragma in LOAD_PRAGMAS:
+			self.conn.execute(pragma)
+		for ddl in DDL:
+			self.conn.execute(ddl)
+		self.inserts = {
+			t: f"INSERT INTO {t} VALUES ({','.join('?' * len(COLUMNS[t]))})" for t in TABLES
+		}
+		self.buffers: dict[str, list[tuple]] = {t: [] for t in TABLES}
+
+	def load(self, cfg: dict, reader: CorpusReader) -> None:
+		pending = 0
+		for table, row in stream_rows(cfg, reader):
+			self.buffers[table].append(row)
+			pending += 1
+			if pending >= BATCH_ROWS:
+				self.flush_all()
+				pending = 0
+		self.flush_all()
+
+	def flush_all(self) -> None:
+		# TABLES is FK-dependency ordered: parents flush before children.
+		for table in TABLES:
+			buf = self.buffers[table]
+			if buf:
+				self.conn.executemany(self.inserts[table], buf)
+				buf.clear()
+		self.conn.commit()
+
+	def finalize(self) -> None:
+		for ddl in INDEX_DDL:
+			self.conn.execute(ddl)
+		self.conn.commit()
+		self.conn.execute("PRAGMA journal_mode=WAL")
+		self.conn.execute("ANALYZE")
+		self.conn.execute("VACUUM")
+		logger.info("indexes created, WAL enabled, ANALYZE + VACUUM done")
+
+	def validate(self, cfg: dict) -> dict[str, int]:
+		counts: dict[str, int] = {}
+		for table, exp in expected_counts(cfg).items():
+			got = self.conn.execute(f"SELECT COUNT(*) FROM {table}").fetchone()[0]
+			counts[table] = got
+			if got != exp:
+				raise ValidationError(f"row count mismatch: {table}: loaded {got} != expected {exp}")
+		reserve = self.conn.execute("SELECT COUNT(*) FROM coupons WHERE run_id IS NULL").fetchone()[0]
+		if reserve != cfg["friction_assignment"]["reserve_coupons_total"]:
+			raise ValidationError(f"reserve coupons: {reserve} != expected")
+		fk_violations = self.conn.execute("PRAGMA foreign_key_check").fetchall()
+		if fk_violations:
+			raise ValidationError(f"foreign_key_check reported {len(fk_violations)} violations")
+		logger.info("all %d table counts match, %d reserve coupons, foreign_key_check clean", len(counts), reserve)
+		return counts
+
+
+def main() -> int:
+	args = parse_task_args("Task 05: convert the CSV corpus to SQLite")
+	out_root: Path = args.out_root
+	db_path = out_root / "sqlite" / "tribo.db"
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		for stale in (db_path, db_path.with_suffix(".db-wal"), db_path.with_suffix(".db-shm")):
+			if stale.exists():
+				logger.info("re-run: removing %s", stale)
+				stale.unlink()
+		cfg = load_lab_config(out_root)
+
+		loader = SqliteLoader(db_path)
+		loader.load(cfg, CorpusReader(out_root / "csv"))
+		loader.finalize()
+		counts = loader.validate(cfg)
+		loader.conn.close()
+
+		db_bytes = db_path.stat().st_size
+		update_storage_sizes(out_root / "bench", "sqlite", [("db", db_bytes)])
+		entries = {
+			"db_file": db_path.as_posix(),
+			"db_bytes": db_bytes,
+			"tables": len(counts),
+			"total_rows": sum(counts.values()),
+			"friction_cycles": counts["friction_cycles"],
+			"friction_loop_points": counts["friction_loop_points"],
+			"tracks": counts["tracks"],
+			"track_summary": counts["track_summary"],
+		}
+		entries.update(process_metrics())
+		marker_path = write_marker(out_root, TASK_ID, entries)
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	print(f"task 05 ok: {db_path} ({db_bytes / (1024 * 1024):.1f} MiB, {len(counts)} tables, {sum(counts.values())} rows)")
+	print(f"bulk rows: cycles={counts['friction_cycles']} loop_points={counts['friction_loop_points']} "
+		f"xrf={counts['xrf_points']} track_summary={counts['track_summary']}")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

docs/Initial_Prompt.md

@@ -0,0 +1,90 @@
+# Prompt: Generate a Multi-File Specification for a Tribology Data Storage Evaluation Pipeline
+
+You are an AI assistant generating a specification for execution by Anthropic Claude Code
+in a Cursor development environment (local filesystem, on-premise only, no cloud).
+
+## Objective
+Produce 12 Markdown files — `00_PLAN.md` plus task files `01`–`11` — that specify a complete,
+reproducible pipeline: simulate a tribology laboratory, generate ~150 MB of realistic
+measurement data in CSV, convert it into 4 target storage formats, benchmark 7 retrieval
+scenarios across all 5 formats with resource metering, extrapolate results to 600 GB+,
+and produce a decision report with tables, charts, and weighted scoring.
+
+## File structure requirements
+Each task file must contain: Context/Goal, Dependencies, Input, Processing instructions,
+Output (artifacts + completion marker `./out/.done/<task>.ok`).
+`00_PLAN.md` must contain: goals, execution environment, a Mermaid dependency flowchart,
+parallelism notes, execution order, and global conventions (ISO-8601 UTC timestamps,
+SI-explicit column names, fixed random seed 20260711, Python with tabs for indentation).
+
+## Laboratory configuration (task 01)
+- Material: Ti-6Al-4V coupons 10×10×3 mm, Cr adhesion layer, Pt-Au coating sputtered in a
+  composition gradient; film thickness 0.3–1.1 µm.
+- Instruments: RAPID custom high-throughput parallelized 6-probe tribometer (friction);
+  Bruker TI980 TriboIndenter (nanoindentation); Bruker M4 Tornado micro-XRF (composition
+  mapping); Kurt J. Lesker PVD 200 sputter-down (deposition); AFM (roughness); optical
+  profilometry (thickness); SIMTRA (sputter-transport Monte-Carlo simulation).
+- Deposition matrix, 4 batches (B721–B724) with gun tilt (20°:0°, 20°:20°, 20°:20°, 0°:20°),
+  Pt/Au power (150/50, 100/100, 150/50, 50/150 W) and discharge voltages.
+- Hierarchy and volumes: batch(×4) → wafer(×3) → coupon(×49, 7×7 grid) = 588 coupons;
+  480 friction coupons (4 RAPID runs × 5 plates × 6 probes × 4 coupons/square),
+  108 reserve; 2 environments (Lab Air / Dry N2, 2 runs each); 3 replicate tracks per
+  coupon with counterface rotation → 1,440 tracks × 1,000 cycles = 1.44 M cycle rows;
+  friction loops (200 points every 100th cycle) → 2.88 M loop points.
+- Physical models for simulation: exponential COF run-in with steady-state depending on
+  Au% and environment; hardness/modulus linear in Au% with noise; log-normal roughness;
+  Archard wear; Au gradient across wafer position (batch-dependent center: 8/25/10/60 wt%).
+
+## Process flow diagrams (task 02) — Mermaid
+Coupon assembly (clean → sonicate → Cr adhesive → PVD sputter); characterization &
+batch assembly (SIMTRA, profilometry, ×49/wafer, ×3/batch, batches 721–724); testing tree
+(friction µ_normal/µ_dry_nit, nanoindentation, AFM, XRF); tribometer session sequence
+(request → samples → ball holders, 3 counterfaces/holder → Excel test plan → setup →
+run → software stop → save avg files → teardown); execution loop (5 plates × 6 probes ×
+4 coupons × 3 tracks with counterface rotation); data-hierarchy ERD.
+
+## Data generation (task 03)
+Hierarchical CSV file tree mirroring the lab structure (batches, simtra, per-coupon
+xrf_map/profilometry/nanoindentation/afm, per-track track_info/cof_vs_cycle/
+friction_loops/wear), streaming generation, MANIFEST.csv with row/byte/sha256,
+size guard 120–200 MB.
+
+## Conversions (tasks 04–07), each recording sizes into storage_sizes.csv
+- JSON-LD: vocab + QUDT + PROV-O; two variants — FULL (bulk embedded, per-coupon files,
+  compact) and HYBRID (metadata only, sourceFile links to CSV).
+- SQLite: single file, integer surrogate keys, WITHOUT ROWID bulk tables, precomputed
+  track_summary, indexes for the benchmark queries, load PRAGMAs, VACUUM/ANALYZE.
+- PostgreSQL 16: same logical schema, RANGE partitioning of bulk tables, COPY loading,
+  b-tree + BRIN indexes, materialized track_summary, pg_dump archive.
+- RDF: N-Triples.gz + Turtle serializations; queryable embedded store (oxigraph preferred,
+  Fuseki/TDB2 fallback); document triple-modeling decisions; expect 25–45 M triples.
+
+## Benchmarks (tasks 08–09)
+Seven scenarios: Q1 single-track COF curve; Q2 steady-state COF at Au=10±0.5 wt%;
+Q3 hardness vs COF join; Q4 condition filter (Dry N2, 100 mN, cof>0.20); Q5 mean run-in
+per batch aggregating ALL raw cycles (summary tables forbidden); Q6 wear volume vs load
+join; Q7 Stribeck-style grouped aggregation. Implement per format (direct path read /
+ijson streaming / SQL / SQL+EXPLAIN / SPARQL); identical result sets validated by checksum.
+Harness: each cell in a separate subprocess; psutil metering (wall time, peak RSS, CPU
+time/utilization, read bytes); 1 warm-up + 3 measured runs, medians; 30-min timeout;
+outputs results_raw.csv / results_median.csv.
+
+## Extrapolation (task 10)
+Project storage linearly and query time by access-pattern complexity laws
+(O(1)/O(log n)/O(n)/single-threaded vs parallel partitioned scan) from the 150 MB
+calibration point to 600 GB, 1.2 TB, 6 TB; flag >1 h IMPRACTICAL, >24 h FAIL; RAM models
+per format; hardware sizing and cost table driven by an editable hw_prices.yaml
+(June 2026 defaults: DDR5 RDIMM ~$14/GB, enterprise NVMe ~$250/TB, 16-core 1U ~$6k).
+
+## Reporting (task 11)
+Tables (footprint + coefficients vs CSV, measured metrics, projected times with flags,
+hardware costs, scoring matrix); matplotlib charts (log scales): footprint, RAM,
+per-query bars, degradation curves 150 MB→6 TB, stacked weighted score, cost-vs-performance
+scatter. Weighted scoring: search speed ×5 (0–50), RAM economy ×2 (0–20), disk economy
+×1 (0–10), max 80; search score from inverse geometric mean of Q1–Q7 projected at 600 GB.
+Concise English narrative: executive summary and use-case mapping (analysis, reports,
+search/filter, archiving, inter-lab exchange), limitations.
+
+## Style
+Specification prose in English, concise and directive. No implementation code in the
+spec — only precise instructions an autonomous coding agent can execute.

docs/research/Tribology_FK_Architecture.html

@@ -0,0 +1,571 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Foreign-Key Architecture Study - Tribology Lab Data Storage</title>
+<link rel="preconnect" href="https://fonts.googleapis.com">
+<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+<link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500;600&family=IBM+Plex+Sans:wght@400;500;600;700&display=swap" rel="stylesheet">
+<style>
+  *{box-sizing:border-box}
+  html,body{margin:0;padding:0}
+  body{font-family:"IBM Plex Sans",system-ui,sans-serif;color:#16181d;background:#eef0f2;-webkit-font-smoothing:antialiased}
+  ::-webkit-scrollbar{width:10px;height:10px}
+  ::-webkit-scrollbar-thumb{background:#cfd4da;border-radius:6px;border:2px solid #eef0f2}
+  ::-webkit-scrollbar-track{background:transparent}
+  .mono{font-family:"IBM Plex Mono",monospace}
+
+  .app{display:flex;flex-direction:column;height:100vh;width:100%;overflow:hidden;background:#eef0f2}
+  header.top{display:flex;align-items:center;gap:14px;height:56px;flex:none;padding:0 22px;background:#fff;border-bottom:1px solid #e2e5ea}
+  .logo{display:flex;align-items:center;justify-content:center;width:30px;height:30px;border-radius:7px;background:#16181d;color:#fff;font-family:"IBM Plex Mono",monospace;font-weight:600;font-size:12px}
+  .htitle{display:flex;flex-direction:column;line-height:1.25;white-space:nowrap}
+  .htitle b{font-size:14.5px;font-weight:600;letter-spacing:-.01em}
+  .htitle span{font-size:11.5px;color:#838a95}
+  .hright{margin-left:auto;display:flex;align-items:center;gap:8px}
+  .chip{font-family:"IBM Plex Mono",monospace;font-size:11px;color:#838a95;background:#f3f5f7;border:1px solid #e6e8ec;padding:4px 10px;border-radius:6px}
+  .chip-ok{display:inline-flex;align-items:center;gap:6px;font-size:12px;color:#0f7a45;background:#e7f6ee;padding:4px 10px;border-radius:6px;font-weight:600}
+  .chip-ok i{width:7px;height:7px;border-radius:7px;background:#18a05a}
+
+  .frame{display:flex;flex:1;min-height:0}
+  nav.side{width:248px;flex:none;background:#fff;border-right:1px solid #e2e5ea;padding:14px 12px;display:flex;flex-direction:column;overflow-y:auto}
+  .navcap{font-family:"IBM Plex Mono",monospace;font-size:10px;letter-spacing:.1em;color:#a4abb4;padding:6px 11px 8px}
+  nav.side a{display:flex;align-items:center;gap:11px;width:100%;padding:8px 11px;border-radius:8px;cursor:pointer;text-decoration:none;font-size:13.5px;margin:1px 0;color:#475059;font-weight:500}
+  nav.side a .tag{width:24px;height:19px;border-radius:5px;display:flex;align-items:center;justify-content:center;font-family:"IBM Plex Mono",monospace;font-size:9px;font-weight:600;flex:none;background:#eceef1;color:#8a909b}
+  nav.side a.active{background:#eaf1fe;color:#16181d;font-weight:600}
+  nav.side a.active .tag{background:#1f5fdb;color:#fff}
+  .sidefoot{margin-top:auto;padding:14px 11px 6px;border-top:1px solid #eef0f2}
+  .sidefoot .cap{font-family:"IBM Plex Mono",monospace;font-size:10px;letter-spacing:.08em;color:#a4abb4;margin-bottom:8px}
+  .sidefoot .txt{font-size:12px;color:#6b727c;line-height:1.6}
+
+  main{flex:1;min-width:0;overflow-y:auto;padding:26px 30px 60px;scroll-behavior:smooth}
+  section{margin-bottom:44px;scroll-margin-top:10px}
+  h1{margin:0;font-size:21px;font-weight:600;letter-spacing:-.02em}
+  h2{margin:0 0 4px;font-size:18px;font-weight:600;letter-spacing:-.02em}
+  .sub{margin:4px 0 18px;font-size:13px;color:#838a95;max-width:860px;line-height:1.55}
+  .card{background:#fff;border:1px solid #e6e8ec;border-radius:11px;padding:18px 20px;margin-bottom:14px}
+  .card h3{margin:0 0 10px;font-size:14px;font-weight:600}
+  .card p{margin:0 0 10px;font-size:13px;color:#3d424b;line-height:1.6;max-width:880px}
+  .card p:last-child{margin-bottom:0}
+  .card ul{margin:6px 0 4px;padding-left:20px}
+  .card li{font-size:13px;color:#3d424b;line-height:1.65;margin-bottom:6px;max-width:840px}
+  .muted{color:#838a95}
+  .kpis{display:grid;grid-template-columns:repeat(4,1fr);gap:14px;margin-bottom:16px}
+  .kpi{background:#fff;border:1px solid #e6e8ec;border-radius:10px;padding:15px 16px}
+  .kpi .l{font-size:11.5px;color:#838a95;font-weight:500;text-transform:uppercase;letter-spacing:.04em}
+  .kpi .v{font-family:"IBM Plex Mono",monospace;font-size:26px;font-weight:600;margin-top:7px;line-height:1}
+  .kpi .s{font-size:11.5px;color:#a4abb4;margin-top:5px}
+  .v-bad{color:#b23b4e}.v-good{color:#0f7a45}.v-neutral{color:#16181d}
+
+  table.data{width:100%;border-collapse:collapse}
+  table.data th{font-size:10.5px;text-transform:uppercase;letter-spacing:.04em;color:#838a95;font-weight:600;text-align:left;padding:9px 12px;background:#f7f8fa;border-bottom:1px solid #eef0f2}
+  table.data td{font-size:12.5px;color:#3d424b;padding:11px 12px;border-bottom:1px solid #f4f5f7;vertical-align:top}
+  table.data td.mono, table.data th.num{font-family:"IBM Plex Mono",monospace}
+  table.data th.num, table.data td.num{text-align:right;font-family:"IBM Plex Mono",monospace}
+  .delta-bad{color:#b23b4e;font-weight:600}
+  .delta-ok{color:#0f7a45;font-weight:600}
+  .delta-zero{color:#838a95}
+
+  pre.code{background:#16181d;color:#dfe3e9;border-radius:9px;padding:14px 16px;font-family:"IBM Plex Mono",monospace;font-size:12px;line-height:1.65;overflow-x:auto;margin:12px 0 0}
+  pre.code .c{color:#7e8794}
+  pre.code .k{color:#7fb1ff}
+  pre.code .hl{color:#ff9db0}
+
+  .erd-wrap{overflow-x:auto;padding:8px 0}
+  .erd{position:relative;margin:0 auto}
+  .erd svg.wire{position:absolute;inset:0;pointer-events:none}
+  .tbl{position:absolute;background:#fff;border:1px solid #d4dae2;border-radius:9px;box-shadow:0 1px 4px rgba(20,30,50,.05);overflow:hidden}
+  .tbl.bulk{border:2px solid #1f5fdb;box-shadow:0 3px 12px rgba(31,95,219,.12)}
+  .tbl-h{background:#16181d;color:#fff;padding:9px 13px;font-family:"IBM Plex Mono",monospace;font-size:12px;font-weight:600;display:flex;justify-content:space-between;gap:8px}
+  .tbl.bulk .tbl-h{background:#1f5fdb}
+  .tbl-h .sub2{color:#7e8794;font-weight:400}
+  .tbl.bulk .tbl-h .sub2{color:#bcd2f7}
+  .tr{display:flex;justify-content:space-between;gap:8px;padding:6px 13px;font-size:11.5px;color:#3d424b;align-items:center}
+  .tr .f{font-family:"IBM Plex Mono",monospace}
+  .tr .t{font-family:"IBM Plex Mono",monospace;font-weight:600;font-size:10.5px;white-space:nowrap}
+  .tr.pk{background:#eaf1fe}.tr.pk .t{color:#1f5fdb}
+  .tr.fk{background:#fbf2e8}.tr.fk .t{color:#c08428}
+  .tr.ref .t{color:#9aa1ab}
+  .tr.idx{background:#fbe9ec}.tr.idx .t{color:#b23b4e}
+  .tr .ty{color:#a4abb4}
+  .notecard{position:absolute;background:#f7f8fa;border:1px dashed #c9cfd8;border-radius:9px;padding:12px 14px;font-size:11.5px;color:#5b6472;line-height:1.55}
+  .notecard b{font-family:"IBM Plex Mono",monospace;font-size:11px;color:#3d424b}
+
+  .hbars{display:flex;flex-direction:column;gap:12px;margin-top:6px}
+  .hbar{display:flex;align-items:center;gap:14px}
+  .hbar .lab{width:250px;flex:none;font-size:12.5px;color:#3d424b}
+  .hbar .lab small{display:block;color:#a4abb4;font-size:11px}
+  .hbar .trk{flex:1;background:#f1f3f6;border-radius:6px;overflow:hidden}
+  .hbar .fill{height:14px;border-radius:6px}
+  .hbar .val{font-family:"IBM Plex Mono",monospace;font-size:12.5px;color:#16181d;width:120px;flex:none;text-align:right}
+  .col-a{background:#9aa6b6}.col-b{background:#1f5fdb}.col-c{background:#c0476b}
+  .legend{display:flex;gap:16px;flex-wrap:wrap;margin:2px 0 10px}
+  .legend span{display:inline-flex;align-items:center;gap:6px;font-size:11.5px;color:#5b6472}
+  .legend i{width:10px;height:10px;border-radius:3px;display:inline-block}
+
+  .callout{border-left:3px solid #1f5fdb;background:#f5f8fe;border-radius:0 9px 9px 0;padding:12px 16px;font-size:13px;color:#2b3038;line-height:1.6;margin:12px 0;max-width:880px}
+  .callout.warn{border-left-color:#b23b4e;background:#fdf3f5}
+  .pill{display:inline-flex;align-items:center;gap:6px;padding:2px 9px;border-radius:20px;font-size:11px;font-weight:600}
+  .pill.rec{color:#0f7a45;background:#e7f6ee}
+  .pill.base{color:#5b6472;background:#eef0f2}
+  .pill.costly{color:#b23b4e;background:#fbe9ec}
+  .foot{font-size:11.5px;color:#a4abb4;border-top:1px solid #e6e8ec;padding-top:14px;margin-top:30px;line-height:1.7}
+</style>
+</head>
+<body>
+<div class="app">
+
+  <header class="top">
+    <div class="logo">FK</div>
+    <div class="htitle">
+      <b>Foreign-Key Architecture Study</b>
+      <span>Tribology lab data storage &middot; bulk-table key design</span>
+    </div>
+    <div class="hright">
+      <span class="chip">SQLite &middot; 4.32M rows measured</span>
+      <span class="chip-ok"><i></i>Measured 2026-07-11</span>
+    </div>
+  </header>
+
+  <div class="frame">
+
+    <nav class="side" id="sidenav">
+      <div class="navcap">NAVIGATE</div>
+      <a href="#overview" class="active"><span class="tag">OV</span><span>Overview</span></a>
+      <a href="#dataset"><span class="tag">DS</span><span>Sample Dataset</span></a>
+      <a href="#variant-a"><span class="tag">A</span><span>Variant A &middot; Composite PK</span></a>
+      <a href="#variant-b"><span class="tag">B</span><span>Variant B &middot; Enforced FK</span></a>
+      <a href="#variant-c"><span class="tag">C</span><span>Variant C &middot; Surrogate id</span></a>
+      <a href="#method"><span class="tag">BM</span><span>Benchmark Method</span></a>
+      <a href="#storage"><span class="tag">ST</span><span>Storage &amp; Load</span></a>
+      <a href="#reads"><span class="tag">RD</span><span>Read Performance</span></a>
+      <a href="#analysis"><span class="tag">AN</span><span>Why It Differs</span></a>
+      <a href="#verdict"><span class="tag">RC</span><span>Recommendation</span></a>
+      <div class="sidefoot">
+        <div class="cap">CONTEXT</div>
+        <div class="txt">Key-schema study for the bulk measurement tables of the LabDataStorageEvaluation pipeline (SQLite / PostgreSQL benchmark targets).</div>
+      </div>
+    </nav>
+
+    <main id="main">
+
+      <!-- ================= OVERVIEW ================= -->
+      <section id="overview">
+        <h1>Overview</h1>
+        <p class="sub">Three candidate key schemas for the two largest measurement tables (<span class="mono">friction_cycles</span>, 1.44M rows and <span class="mono">friction_loop_points</span>, 2.88M rows) were built from the same simulated corpus and measured for disk footprint, load time and six read patterns. In every variant the <b>relations between tables are single-column integer foreign keys</b> (<span class="mono">track_id</span>, <span class="mono">coupon_id</span>); the variants differ only in how a bulk row is identified inside its own table.</p>
+
+        <div class="kpis">
+          <div class="kpi"><div class="l">Disk, variant C vs A</div><div class="v v-bad">+74%</div><div class="s">190.2 vs 109.1 MiB</div></div>
+          <div class="kpi"><div class="l">GROUP BY scan, C vs A</div><div class="v v-bad">+59%</div><div class="s">135.9 vs 85.7 ms (Q5 pattern)</div></div>
+          <div class="kpi"><div class="l">Enforced FK cost (B)</div><div class="v v-good">+0.5 s</div><div class="s">load only; disk and reads = A</div></div>
+          <div class="kpi"><div class="l">Point lookups</div><div class="v v-neutral">&plusmn;2%</div><div class="s">identical across A / B / C</div></div>
+        </div>
+
+        <div class="card">
+          <h3>The three variants</h3>
+          <ul>
+            <li><b>Variant A - composite natural PK.</b> Row identity of a bulk row is its natural composite key <span class="mono">(track_id, cycle)</span>, stored as a clustered <span class="mono">WITHOUT ROWID</span> table. FK columns are plain integers, validated by the loader against the corpus manifest (no engine constraint).</li>
+            <li><b>Variant B - composite natural PK + enforced FK.</b> Same physical layout as A, plus <span class="mono">REFERENCES</span> constraints with <span class="mono">PRAGMA foreign_keys=ON</span>, so the engine itself guarantees referential integrity.</li>
+            <li><b>Variant C - surrogate key on every table.</b> Classic style: each table gets a single-column auto-increment key named <span class="mono">&lt;singular&gt;_id</span> (<span class="mono">friction_cycle_id</span>, never a bare <span class="mono">id</span>). Correctness then requires an additional <span class="mono">UNIQUE(track_id, cycle)</span> index - a second B-tree.</li>
+          </ul>
+          <div class="callout">Referential integrity is <b>not</b> what differs between the variants: every relation is a single integer FK in all three. What differs is the number of B-trees the engine must store and traverse per bulk table: one (A, B) or two (C).</div>
+        </div>
+      </section>
+
+      <!-- ================= DATASET ================= -->
+      <section id="dataset">
+        <h2>Sample Dataset</h2>
+        <p class="sub">All measurements ran against the real generated corpus of the evaluation pipeline - deterministic, physically plausible tribology data (seed 20260711). The two bulk tables and their parent dimension were loaded in full.</p>
+        <div class="card" style="padding:0;overflow:hidden">
+          <table class="data">
+            <thead><tr><th>Table</th><th>Grain</th><th class="num">Rows loaded</th><th>Columns (beyond keys)</th><th>Source files</th></tr></thead>
+            <tbody>
+              <tr><td class="mono">tracks</td><td>one friction track (3 per tested coupon)</td><td class="num">1,440</td><td>track_code, coupon_id, run_id, environment, load_mN, ...</td><td class="mono">track_info.csv &times; 1,440</td></tr>
+              <tr><td class="mono">friction_cycles</td><td>one reciprocating cycle of one track</td><td class="num">1,440,000</td><td>cof</td><td class="mono">cof_vs_cycle.csv &times; 1,440</td></tr>
+              <tr><td class="mono">friction_loop_points</td><td>one position-resolved loop point (10 loops &times; 200 pts per track)</td><td class="num">2,880,000</td><td>position_um, friction_force_mN</td><td class="mono">friction_loops.csv &times; 1,440</td></tr>
+            </tbody>
+          </table>
+        </div>
+        <div class="card">
+          <h3>Corpus context</h3>
+          <p>The corpus models a Pt-Au coating study: 4 deposition batches &rarr; 12 wafers &rarr; 588 coupons (480 friction-tested) &rarr; 1,440 tracks &rarr; 1.44M cycle records. Generated size: 141.8 MiB of CSV; the production target for extrapolation is 600 GB - 6 TB, which is why per-row bytes on the bulk tables matter.</p>
+        </div>
+      </section>
+
+      <!-- ================= VARIANT A ================= -->
+      <section id="variant-a">
+        <h2>Variant A &middot; Composite Natural Primary Key <span class="pill base" style="vertical-align:2px">baseline</span></h2>
+        <p class="sub">The bulk row is identified by what it physically is: cycle <span class="mono">n</span> of track <span class="mono">t</span>. The composite PK doubles as the clustering order, so the table itself is the only B-tree. FK columns are unenforced integers; integrity is guaranteed by row-count and checksum validation against the corpus manifest.</p>
+
+        <div class="card">
+          <div class="erd-wrap"><div class="erd" style="width:1000px;height:440px">
+            <svg class="wire" viewBox="0 0 1000 440" width="1000" height="440">
+              <path d="M270 200 H360 V77 H450" stroke="#c5cdd8" stroke-width="1.5" fill="none"/>
+              <path d="M270 240 H360 V269 H450" stroke="#c5cdd8" stroke-width="1.5" fill="none"/>
+              <circle cx="270" cy="200" r="3.5" fill="#1f5fdb"/><circle cx="270" cy="240" r="3.5" fill="#1f5fdb"/>
+              <text x="280" y="193" font-family="IBM Plex Mono, monospace" font-size="11" fill="#1f5fdb">1</text>
+              <text x="280" y="233" font-family="IBM Plex Mono, monospace" font-size="11" fill="#1f5fdb">1</text>
+              <text x="432" y="70" font-family="IBM Plex Mono, monospace" font-size="11" fill="#9aa1ab">&#8734;</text>
+              <text x="432" y="262" font-family="IBM Plex Mono, monospace" font-size="11" fill="#9aa1ab">&#8734;</text>
+            </svg>
+            <div class="tbl" style="left:20px;top:120px;width:250px">
+              <div class="tbl-h"><span>tracks</span><span class="sub2">dimension</span></div>
+              <div class="tr pk"><span class="f">track_id</span><span class="t">PK</span></div>
+              <div class="tr"><span class="f">track_code</span><span class="t" style="color:#1f5fdb">UK</span></div>
+              <div class="tr fk"><span class="f">coupon_id</span><span class="t">FK</span></div>
+              <div class="tr fk"><span class="f">run_id</span><span class="t">FK</span></div>
+              <div class="tr"><span class="f">environment</span><span class="ty">text</span></div>
+              <div class="tr"><span class="f">load_mn</span><span class="ty">real</span></div>
+            </div>
+            <div class="tbl bulk" style="left:450px;top:20px;width:280px">
+              <div class="tbl-h"><span>friction_cycles</span><span class="sub2">1.44M rows</span></div>
+              <div class="tr pk"><span class="f">track_id</span><span class="t">PK &middot; ref</span></div>
+              <div class="tr pk"><span class="f">cycle</span><span class="t">PK</span></div>
+              <div class="tr"><span class="f">cof</span><span class="ty">real</span></div>
+            </div>
+            <div class="tbl bulk" style="left:450px;top:185px;width:280px">
+              <div class="tbl-h"><span>friction_loop_points</span><span class="sub2">2.88M rows</span></div>
+              <div class="tr pk"><span class="f">track_id</span><span class="t">PK &middot; ref</span></div>
+              <div class="tr pk"><span class="f">cycle</span><span class="t">PK</span></div>
+              <div class="tr pk"><span class="f">pt</span><span class="t">PK</span></div>
+              <div class="tr"><span class="f">position_um</span><span class="ty">real</span></div>
+              <div class="tr"><span class="f">friction_force_mn</span><span class="ty">real</span></div>
+            </div>
+            <div class="notecard" style="left:770px;top:130px;width:210px">
+              <b>wear</b> and <b>track_summary</b> follow the same pattern: single-column PK = FK <b>track_id</b> (1:1 with tracks).<br><br>
+              <b>ref</b> = integer reference to tracks(track_id), validated by the loader, no engine constraint.
+            </div>
+          </div></div>
+          <pre class="code">CREATE TABLE friction_cycles (
+  track_id INTEGER NOT NULL,   <span class="c">-- reference to tracks(track_id), loader-validated</span>
+  cycle    INTEGER NOT NULL,
+  cof      REAL    NOT NULL,
+  <span class="k">PRIMARY KEY (track_id, cycle)</span>
+) <span class="k">WITHOUT ROWID</span>;              <span class="c">-- clustered: the PK is the table's only B-tree</span></pre>
+        </div>
+      </section>
+
+      <!-- ================= VARIANT B ================= -->
+      <section id="variant-b">
+        <h2>Variant B &middot; Composite PK + Enforced Foreign Keys <span class="pill rec" style="vertical-align:2px">recommended</span></h2>
+        <p class="sub">Physically identical to Variant A - same clustered layout, same single B-tree per bulk table. The only change: FK columns carry real <span class="mono">REFERENCES</span> constraints and every connection runs <span class="mono">PRAGMA foreign_keys=ON</span>, so the DBMS itself rejects an orphan row. Measured cost: +0.5 s of load time on 4.32M rows; zero cost on disk and reads.</p>
+
+        <div class="card">
+          <div class="erd-wrap"><div class="erd" style="width:1000px;height:440px">
+            <svg class="wire" viewBox="0 0 1000 440" width="1000" height="440">
+              <path d="M270 200 H360 V77 H450" stroke="#c5cdd8" stroke-width="1.5" fill="none"/>
+              <path d="M270 240 H360 V269 H450" stroke="#c5cdd8" stroke-width="1.5" fill="none"/>
+              <circle cx="270" cy="200" r="3.5" fill="#1f5fdb"/><circle cx="270" cy="240" r="3.5" fill="#1f5fdb"/>
+              <text x="280" y="193" font-family="IBM Plex Mono, monospace" font-size="11" fill="#1f5fdb">1</text>
+              <text x="280" y="233" font-family="IBM Plex Mono, monospace" font-size="11" fill="#1f5fdb">1</text>
+              <text x="432" y="70" font-family="IBM Plex Mono, monospace" font-size="11" fill="#9aa1ab">&#8734;</text>
+              <text x="432" y="262" font-family="IBM Plex Mono, monospace" font-size="11" fill="#9aa1ab">&#8734;</text>
+            </svg>
+            <div class="tbl" style="left:20px;top:120px;width:250px">
+              <div class="tbl-h"><span>tracks</span><span class="sub2">dimension</span></div>
+              <div class="tr pk"><span class="f">track_id</span><span class="t">PK</span></div>
+              <div class="tr"><span class="f">track_code</span><span class="t" style="color:#1f5fdb">UK</span></div>
+              <div class="tr fk"><span class="f">coupon_id</span><span class="t">FK</span></div>
+              <div class="tr fk"><span class="f">run_id</span><span class="t">FK</span></div>
+              <div class="tr"><span class="f">environment</span><span class="ty">text</span></div>
+              <div class="tr"><span class="f">load_mn</span><span class="ty">real</span></div>
+            </div>
+            <div class="tbl bulk" style="left:450px;top:20px;width:280px">
+              <div class="tbl-h"><span>friction_cycles</span><span class="sub2">1.44M rows</span></div>
+              <div class="tr pk"><span class="f">track_id</span><span class="t">PK</span></div>
+              <div class="tr fk"><span class="f">track_id &rarr; tracks</span><span class="t">FK &#10003;</span></div>
+              <div class="tr pk"><span class="f">cycle</span><span class="t">PK</span></div>
+              <div class="tr"><span class="f">cof</span><span class="ty">real</span></div>
+            </div>
+            <div class="tbl bulk" style="left:450px;top:190px;width:280px">
+              <div class="tbl-h"><span>friction_loop_points</span><span class="sub2">2.88M rows</span></div>
+              <div class="tr pk"><span class="f">track_id</span><span class="t">PK</span></div>
+              <div class="tr fk"><span class="f">track_id &rarr; tracks</span><span class="t">FK &#10003;</span></div>
+              <div class="tr pk"><span class="f">cycle</span><span class="t">PK</span></div>
+              <div class="tr pk"><span class="f">pt</span><span class="t">PK</span></div>
+              <div class="tr"><span class="f">position_um, friction_force_mn</span><span class="ty">real</span></div>
+            </div>
+            <div class="notecard" style="left:770px;top:130px;width:210px">
+              <b>PRAGMA foreign_keys = ON</b> on every connection - SQLite silently ignores FK clauses without it.<br><br>
+              FK checks are lookups into the already-indexed parent PK, which is why the measured overhead is only ~6% of load time.
+            </div>
+          </div></div>
+          <pre class="code">PRAGMA foreign_keys = ON;
+
+CREATE TABLE friction_cycles (
+  track_id INTEGER NOT NULL <span class="k">REFERENCES tracks(track_id)</span>,
+  cycle    INTEGER NOT NULL,
+  cof      REAL    NOT NULL,
+  <span class="k">PRIMARY KEY (track_id, cycle)</span>
+) <span class="k">WITHOUT ROWID</span>;</pre>
+        </div>
+      </section>
+
+      <!-- ================= VARIANT C ================= -->
+      <section id="variant-c">
+        <h2>Variant C &middot; Surrogate <span class="mono" style="font-size:16px">&lt;singular&gt;_id</span> on Every Table <span class="pill costly" style="vertical-align:2px">+74% disk</span></h2>
+        <p class="sub">The uniform classic design: every table, including the bulk fact tables, gets a single-column auto-increment key (<span class="mono">friction_cycle_id</span>, <span class="mono">friction_loop_point_id</span> - by convention always <span class="mono">&lt;table name in singular&gt;_id</span>, never a bare <span class="mono">id</span>). In SQLite the id column itself is nearly free (it aliases the rowid), but correctness still demands <span class="mono">UNIQUE(track_id, cycle)</span> - and that unique index is a second B-tree holding a copy of almost every column.</p>
+
+        <div class="card">
+          <div class="erd-wrap"><div class="erd" style="width:1000px;height:520px">
+            <svg class="wire" viewBox="0 0 1000 520" width="1000" height="520">
+              <path d="M270 240 H360 V100 H450" stroke="#c5cdd8" stroke-width="1.5" fill="none"/>
+              <path d="M270 280 H360 V345 H450" stroke="#c5cdd8" stroke-width="1.5" fill="none"/>
+              <circle cx="270" cy="240" r="3.5" fill="#1f5fdb"/><circle cx="270" cy="280" r="3.5" fill="#1f5fdb"/>
+              <text x="280" y="233" font-family="IBM Plex Mono, monospace" font-size="11" fill="#1f5fdb">1</text>
+              <text x="280" y="273" font-family="IBM Plex Mono, monospace" font-size="11" fill="#1f5fdb">1</text>
+              <text x="432" y="93" font-family="IBM Plex Mono, monospace" font-size="11" fill="#9aa1ab">&#8734;</text>
+              <text x="432" y="338" font-family="IBM Plex Mono, monospace" font-size="11" fill="#9aa1ab">&#8734;</text>
+            </svg>
+            <div class="tbl" style="left:20px;top:160px;width:250px">
+              <div class="tbl-h"><span>tracks</span><span class="sub2">dimension</span></div>
+              <div class="tr pk"><span class="f">track_id</span><span class="t">PK</span></div>
+              <div class="tr"><span class="f">track_code</span><span class="t" style="color:#1f5fdb">UK</span></div>
+              <div class="tr fk"><span class="f">coupon_id</span><span class="t">FK</span></div>
+              <div class="tr fk"><span class="f">run_id</span><span class="t">FK</span></div>
+              <div class="tr"><span class="f">environment</span><span class="ty">text</span></div>
+              <div class="tr"><span class="f">load_mn</span><span class="ty">real</span></div>
+            </div>
+            <div class="tbl bulk" style="left:450px;top:20px;width:280px">
+              <div class="tbl-h"><span>friction_cycles</span><span class="sub2">1.44M rows</span></div>
+              <div class="tr pk"><span class="f">friction_cycle_id</span><span class="t">PK</span></div>
+              <div class="tr fk"><span class="f">track_id &rarr; tracks</span><span class="t">FK &#10003;</span></div>
+              <div class="tr"><span class="f">cycle</span><span class="ty">int</span></div>
+              <div class="tr"><span class="f">cof</span><span class="ty">real</span></div>
+              <div class="tr idx"><span class="f">UNIQUE (track_id, cycle)</span><span class="t">2nd B-tree</span></div>
+            </div>
+            <div class="tbl bulk" style="left:450px;top:235px;width:280px">
+              <div class="tbl-h"><span>friction_loop_points</span><span class="sub2">2.88M rows</span></div>
+              <div class="tr pk"><span class="f">friction_loop_point_id</span><span class="t">PK</span></div>
+              <div class="tr fk"><span class="f">track_id &rarr; tracks</span><span class="t">FK &#10003;</span></div>
+              <div class="tr"><span class="f">cycle</span><span class="ty">int</span></div>
+              <div class="tr"><span class="f">pt</span><span class="ty">int</span></div>
+              <div class="tr"><span class="f">position_um</span><span class="ty">real</span></div>
+              <div class="tr"><span class="f">friction_force_mn</span><span class="ty">real</span></div>
+              <div class="tr idx"><span class="f">UNIQUE (track_id, cycle, pt)</span><span class="t">2nd B-tree</span></div>
+            </div>
+            <div class="notecard" style="left:770px;top:180px;width:210px">
+              The surrogate key is <b>never referenced</b> by any other table - no row in the schema points at an individual cycle.<br><br>
+              The UNIQUE index is <b>mandatory for correctness</b> (it prevents duplicate cycles), so its cost cannot be avoided in this variant.
+            </div>
+          </div></div>
+          <pre class="code">CREATE TABLE friction_cycles (
+  <span class="hl">friction_cycle_id INTEGER PRIMARY KEY</span>,        <span class="c">-- rowid alias; unused by any FK</span>
+  track_id INTEGER NOT NULL REFERENCES tracks(track_id),
+  cycle    INTEGER NOT NULL,
+  cof      REAL    NOT NULL,
+  <span class="hl">UNIQUE (track_id, cycle)</span>                      <span class="c">-- required; duplicates the table</span>
+);</pre>
+        </div>
+      </section>
+
+      <!-- ================= METHOD ================= -->
+      <section id="method">
+        <h2>Benchmark Method</h2>
+        <p class="sub">Identical protocol for all three variants; only the DDL differs.</p>
+        <div class="card">
+          <ul>
+            <li><b>Data:</b> the full generated corpus subset - <span class="mono">tracks</span> (1,440), <span class="mono">friction_cycles</span> (1,440,000), <span class="mono">friction_loop_points</span> (2,880,000) - read from the same CSV files in the same order for every variant.</li>
+            <li><b>Load:</b> SQLite, <span class="mono">PRAGMA journal_mode=OFF, synchronous=OFF, cache_size=-262144</span>; inserts via <span class="mono">executemany</span> in 50k-row batches; <span class="mono">VACUUM</span> before measuring the file size.</li>
+            <li><b>Reads:</b> warm OS page cache, best of 3 runs per pattern; fixed deterministic probe sets (39 tracks for range reads, 150 <span class="mono">(track_id, cycle)</span> pairs for point lookups); result-set sizes asserted, so every variant answers the identical question.</li>
+            <li><b>Caveat:</b> at 150 MB everything fits in RAM, which <i>flatters</i> variant C - at the 600 GB production target the working set exceeds RAM, and twice the pages means twice the cache pressure and disk reads. Warm-cache deltas are therefore a lower bound.</li>
+          </ul>
+        </div>
+      </section>
+
+      <!-- ================= STORAGE ================= -->
+      <section id="storage">
+        <h2>Results &middot; Storage &amp; Load</h2>
+        <p class="sub">File size after VACUUM; load time includes CSV parsing (identical work in all variants).</p>
+
+        <div class="card" style="padding:0;overflow:hidden">
+          <table class="data">
+            <thead><tr><th>Variant</th><th>Bulk-table keys</th><th class="num">Load time (s)</th><th class="num">DB size (MiB)</th><th class="num">Size vs A</th></tr></thead>
+            <tbody>
+              <tr><td><b>A</b> &middot; composite PK</td><td>PK (track_id, cycle), unenforced refs</td><td class="num">7.5</td><td class="num">109.1</td><td class="num delta-zero">baseline</td></tr>
+              <tr><td><b>B</b> &middot; composite PK + FK</td><td>PK (track_id, cycle), REFERENCES enforced</td><td class="num">8.0</td><td class="num">109.1</td><td class="num delta-ok">&plusmn;0%</td></tr>
+              <tr><td><b>C</b> &middot; surrogate id</td><td>friction_cycle_id PK + UNIQUE(track_id, cycle)</td><td class="num">8.8</td><td class="num">190.2</td><td class="num delta-bad">+74%</td></tr>
+            </tbody>
+          </table>
+        </div>
+
+        <div class="card">
+          <h3>Database size after VACUUM (MiB, scale 0-200)</h3>
+          <div class="hbars">
+            <div class="hbar"><span class="lab">Variant A<small>composite PK, unenforced refs</small></span><div class="trk"><div class="fill col-a" style="width:54.6%"></div></div><span class="val">109.1 MiB</span></div>
+            <div class="hbar"><span class="lab">Variant B<small>composite PK + enforced FK</small></span><div class="trk"><div class="fill col-b" style="width:54.6%"></div></div><span class="val">109.1 MiB</span></div>
+            <div class="hbar"><span class="lab">Variant C<small>surrogate id + UNIQUE index</small></span><div class="trk"><div class="fill col-c" style="width:95.1%"></div></div><span class="val">190.2 MiB</span></div>
+          </div>
+          <h3 style="margin-top:20px">Load time, 4.32M rows (s, scale 0-10)</h3>
+          <div class="hbars">
+            <div class="hbar"><span class="lab">Variant A</span><div class="trk"><div class="fill col-a" style="width:75%"></div></div><span class="val">7.5 s</span></div>
+            <div class="hbar"><span class="lab">Variant B</span><div class="trk"><div class="fill col-b" style="width:80%"></div></div><span class="val">8.0 s</span></div>
+            <div class="hbar"><span class="lab">Variant C</span><div class="trk"><div class="fill col-c" style="width:88%"></div></div><span class="val">8.8 s</span></div>
+          </div>
+          <div class="callout" style="margin-top:16px">Extrapolated to the 600 GB production target, the +74% of variant C on the two hottest tables translates into <b>hundreds of gigabytes of extra disk</b> and a doubled page count competing for the same RAM cache.</div>
+        </div>
+      </section>
+
+      <!-- ================= READS ================= -->
+      <section id="reads">
+        <h2>Results &middot; Read Performance</h2>
+        <p class="sub">Warm OS cache, best of 3 runs. Q1 / Q5 refer to the benchmark scenarios of the evaluation pipeline.</p>
+
+        <div class="card" style="padding:0;overflow:hidden">
+          <table class="data">
+            <thead><tr><th>Read pattern</th><th>Unit</th><th class="num">A</th><th class="num">B</th><th class="num">C</th><th class="num">C vs A</th></tr></thead>
+            <tbody>
+              <tr><td>Point lookup of one row by <span class="mono">(track_id, cycle)</span></td><td class="mono">&micro;s/op</td><td class="num">24.8</td><td class="num">23.6</td><td class="num">24.3</td><td class="num delta-zero">~0%</td></tr>
+              <tr><td>Q1: all 1,000 cycles of one track</td><td class="mono">ms/track</td><td class="num">0.473</td><td class="num">0.485</td><td class="num">0.533</td><td class="num delta-bad">+13%</td></tr>
+              <tr><td>All 2,000 loop points of one track</td><td class="mono">ms/track</td><td class="num">1.267</td><td class="num">1.299</td><td class="num">1.336</td><td class="num delta-bad">+5%</td></tr>
+              <tr><td>Full scan + AVG over 1.44M cycles</td><td class="mono">ms</td><td class="num">64.9</td><td class="num">64.9</td><td class="num">70.7</td><td class="num delta-bad">+9%</td></tr>
+              <tr><td>AVG with GROUP BY track_id over 1.44M cycles (Q5 pattern)</td><td class="mono">ms</td><td class="num">85.7</td><td class="num">89.1</td><td class="num">135.9</td><td class="num delta-bad">+59%</td></tr>
+              <tr><td>Full scan over 2.88M loop points</td><td class="mono">ms</td><td class="num">152.5</td><td class="num">150.8</td><td class="num">165.4</td><td class="num delta-bad">+8%</td></tr>
+            </tbody>
+          </table>
+        </div>
+
+        <div class="card">
+          <h3>Relative read time, variant A = 100% (lower is better)</h3>
+          <div class="legend">
+            <span><i class="col-a"></i>A &middot; composite PK</span>
+            <span><i class="col-b"></i>B &middot; + enforced FK</span>
+            <span><i class="col-c"></i>C &middot; surrogate id + UNIQUE</span>
+          </div>
+          <svg viewBox="0 0 1080 300" style="width:100%;height:auto;display:block">
+            <line x1="60" y1="246" x2="1080" y2="246" stroke="#d8dce1" stroke-width="1"/>
+            <line x1="60" y1="183.2" x2="1080" y2="183.2" stroke="#eef0f2" stroke-width="1"/>
+            <line x1="60" y1="120.4" x2="1080" y2="120.4" stroke="#c9d4e6" stroke-width="1" stroke-dasharray="4 3"/>
+            <line x1="60" y1="57.7" x2="1080" y2="57.7" stroke="#eef0f2" stroke-width="1"/>
+            <text x="54" y="246" text-anchor="end" dominant-baseline="middle" font-family="IBM Plex Mono, monospace" font-size="9" fill="#a4abb4">0%</text>
+            <text x="54" y="183.2" text-anchor="end" dominant-baseline="middle" font-family="IBM Plex Mono, monospace" font-size="9" fill="#a4abb4">50%</text>
+            <text x="54" y="120.4" text-anchor="end" dominant-baseline="middle" font-family="IBM Plex Mono, monospace" font-size="9" fill="#7e8ba1">100%</text>
+            <text x="54" y="57.7" text-anchor="end" dominant-baseline="middle" font-family="IBM Plex Mono, monospace" font-size="9" fill="#a4abb4">150%</text>
+
+            <!-- group 0: point lookup -->
+            <rect x="78"  y="120.4" width="38" height="125.6" rx="2.5" fill="#9aa6b6"/>
+            <rect x="124" y="126.5" width="38" height="119.5" rx="2.5" fill="#1f5fdb"/>
+            <rect x="170" y="123.0" width="38" height="123.0" rx="2.5" fill="#c0476b"/>
+            <text x="143" y="262" text-anchor="middle" font-size="10" fill="#5b6472">Point lookup</text>
+            <!-- group 1: Q1 cycles -->
+            <rect x="245" y="120.4" width="38" height="125.6" rx="2.5" fill="#9aa6b6"/>
+            <rect x="291" y="117.3" width="38" height="128.7" rx="2.5" fill="#1f5fdb"/>
+            <rect x="337" y="104.5" width="38" height="141.5" rx="2.5" fill="#c0476b"/>
+            <text x="310" y="262" text-anchor="middle" font-size="10" fill="#5b6472">Q1: track cycles</text>
+            <!-- group 2: loop points range -->
+            <rect x="411" y="120.4" width="38" height="125.6" rx="2.5" fill="#9aa6b6"/>
+            <rect x="457" y="117.3" width="38" height="128.7" rx="2.5" fill="#1f5fdb"/>
+            <rect x="503" y="113.7" width="38" height="132.3" rx="2.5" fill="#c0476b"/>
+            <text x="476" y="262" text-anchor="middle" font-size="10" fill="#5b6472">Track loop points</text>
+            <!-- group 3: full scan avg -->
+            <rect x="578" y="120.4" width="38" height="125.6" rx="2.5" fill="#9aa6b6"/>
+            <rect x="624" y="120.4" width="38" height="125.6" rx="2.5" fill="#1f5fdb"/>
+            <rect x="670" y="109.3" width="38" height="136.7" rx="2.5" fill="#c0476b"/>
+            <text x="643" y="262" text-anchor="middle" font-size="10" fill="#5b6472">Full scan + AVG</text>
+            <!-- group 4: group by -->
+            <rect x="745" y="120.4" width="38" height="125.6" rx="2.5" fill="#9aa6b6"/>
+            <rect x="791" y="115.4" width="38" height="130.6" rx="2.5" fill="#1f5fdb"/>
+            <rect x="837" y="46.9"  width="38" height="199.1" rx="2.5" fill="#c0476b"/>
+            <text x="856" y="40" text-anchor="middle" font-family="IBM Plex Mono, monospace" font-size="10" font-weight="600" fill="#b23b4e">+59%</text>
+            <text x="810" y="262" text-anchor="middle" font-size="10" fill="#5b6472">GROUP BY track (Q5)</text>
+            <!-- group 5: loops scan -->
+            <rect x="911"  y="120.4" width="38" height="125.6" rx="2.5" fill="#9aa6b6"/>
+            <rect x="957"  y="121.8" width="38" height="124.2" rx="2.5" fill="#1f5fdb"/>
+            <rect x="1003" y="109.8" width="38" height="136.2" rx="2.5" fill="#c0476b"/>
+            <text x="976" y="262" text-anchor="middle" font-size="10" fill="#5b6472">Loop points scan</text>
+
+            <text x="570" y="290" text-anchor="middle" font-size="10" fill="#838a95">dashed line = variant A baseline (100%)</text>
+          </svg>
+        </div>
+      </section>
+
+      <!-- ================= ANALYSIS ================= -->
+      <section id="analysis">
+        <h2>Why It Differs</h2>
+        <p class="sub">The whole difference reduces to one thing: how many B-trees per bulk table, and in what order the rows physically live.</p>
+
+        <div class="card">
+          <div class="erd-wrap"><div class="erd" style="width:1000px;height:230px">
+            <svg class="wire" viewBox="0 0 1000 230" width="1000" height="230">
+              <path d="M700 115 H760" stroke="#b23b4e" stroke-width="1.6" fill="none" stroke-dasharray="5 4"/>
+              <text x="730" y="105" text-anchor="middle" font-family="IBM Plex Mono, monospace" font-size="10" fill="#b23b4e">extra hop</text>
+            </svg>
+            <div class="tbl" style="left:20px;top:30px;width:380px;border-color:#0f9d8a">
+              <div class="tbl-h" style="background:#0f9d8a"><span>Variants A / B &middot; one B-tree</span></div>
+              <div class="tr"><span class="f">clustered table keyed by (track_id, cycle)</span></div>
+              <div class="tr"><span class="f">descent &rarr; leaf &rarr; sequential range scan</span></div>
+              <div class="tr"><span class="f muted">rows of one track are physically adjacent</span></div>
+            </div>
+            <div class="tbl" style="left:450px;top:10px;width:250px;border-color:#c0476b">
+              <div class="tbl-h" style="background:#c0476b"><span>Variant C &middot; B-tree 1</span></div>
+              <div class="tr"><span class="f">UNIQUE index (track_id, cycle)</span></div>
+              <div class="tr"><span class="f muted">stores key columns + rowid again</span></div>
+            </div>
+            <div class="tbl" style="left:760px;top:80px;width:220px;border-color:#c0476b">
+              <div class="tbl-h" style="background:#c0476b"><span>Variant C &middot; B-tree 2</span></div>
+              <div class="tr"><span class="f">table keyed by friction_cycle_id</span></div>
+              <div class="tr"><span class="f muted">per-row lookup to fetch cof</span></div>
+            </div>
+          </div></div>
+          <ul>
+            <li><b>+74% disk.</b> In SQLite the surrogate id column is a rowid alias and costs nothing in the table itself - but the mandatory <span class="mono">UNIQUE(track_id, cycle)</span> index is a full second B-tree whose entries contain almost every column of the row. Two trees instead of one on a 3-column table is close to a doubling.</li>
+            <li><b>+59% on grouped aggregation (the pipeline's forced-full-scan Q5).</b> In A/B the table is clustered by <span class="mono">track_id</span>, so <span class="mono">GROUP BY track_id</span> streams the tree in group order. In C the engine walks the UNIQUE index for ordering and hops into the table B-tree for every <span class="mono">cof</span> value - the classic double access path.</li>
+            <li><b>+5-13% on ranges and plain scans.</b> More pages (190 vs 109 MiB) means more page traversals and cell decoding, even fully cached.</li>
+            <li><b>~0% on point lookups.</b> Tree depth is the same either way; a single-row probe does not expose the second tree.</li>
+            <li><b>PostgreSQL outlook.</b> PG has no WITHOUT ROWID: the composite-PK table is a heap plus one PK index either way, but variant C still adds the id column to every heap tuple <i>and</i> the second index - expect +30-50% on the bulk tables, plus higher COPY-time FK validation cost.</li>
+          </ul>
+          <div class="callout warn"><b>Integrity is a constant across all variants.</b> Every relation is a single-column integer FK. Choosing C buys naming uniformity, not more integrity - and the surrogate key it adds is never referenced by anything.</div>
+        </div>
+      </section>
+
+      <!-- ================= RECOMMENDATION ================= -->
+      <section id="verdict">
+        <h2>Recommendation</h2>
+        <div class="card" style="border-left:4px solid #0f7a45">
+          <h3><span class="pill rec">Variant B</span> &nbsp;Composite natural PK on bulk tables + enforced foreign keys</h3>
+          <ul>
+            <li>Referential integrity is guaranteed by the DBMS itself at a measured cost of <b>+0.5 s of load time</b> and nothing else.</li>
+            <li>Disk footprint and read profile are identical to the leanest variant A: <b>109.1 MiB</b>, single clustered B-tree per bulk table.</li>
+            <li>Dimension tables (<span class="mono">batches</span>, <span class="mono">wafers</span>, <span class="mono">coupons</span>, <span class="mono">runs</span>, <span class="mono">tracks</span>, <span class="mono">instruments</span>) keep their classic single-column surrogate keys - <span class="mono">&lt;singular&gt;_id</span>, never a bare <span class="mono">id</span> - because other tables reference their rows.</li>
+            <li>Variant C remains a valid design where fact rows must be individually addressable from outside (annotations, reprocessing links). No table in this schema references an individual cycle or loop point, so its cost (+74% disk, up to +59% on aggregation, growing at 600 GB scale) buys nothing here.</li>
+          </ul>
+        </div>
+        <div class="foot">
+          Measured 2026-07-11 on the deterministic simulated corpus (seed 20260711, 141.8 MiB CSV) &middot; SQLite, warm OS cache, best of 3 runs &middot;
+          scripts: <span class="mono">measure_key_variants.py</span> (build + size + load + Q1) and <span class="mono">read_benchmarks.py</span> (six read patterns) &middot;
+          companion page: <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html" style="color:#1f5fdb;text-decoration:none">Storage Format Evaluation</a> (full study results).
+        </div>
+      </section>
+
+    </main>
+  </div>
+</div>
+
+<script>
+(function(){
+  var links = Array.prototype.slice.call(document.querySelectorAll('#sidenav a[href^="#"]'));
+  var byId = {};
+  links.forEach(function(a){ byId[a.getAttribute('href').slice(1)] = a; });
+  var sections = Array.prototype.slice.call(document.querySelectorAll('main section'));
+  var main = document.getElementById('main');
+  function setActive(id){
+    links.forEach(function(a){ a.classList.remove('active'); });
+    if(byId[id]) byId[id].classList.add('active');
+  }
+  links.forEach(function(a){
+    a.addEventListener('click', function(){ setActive(a.getAttribute('href').slice(1)); });
+  });
+  var ticking = false;
+  main.addEventListener('scroll', function(){
+    if(ticking) return; ticking = true;
+    requestAnimationFrame(function(){
+      var top = main.scrollTop + 90, current = sections[0].id;
+      sections.forEach(function(s){ if(s.offsetTop <= top) current = s.id; });
+      setActive(current);
+      ticking = false;
+    });
+  });
+})();
+</script>
+</body>
+</html>

docs/research/Tribology_Storage_Decision.html

@@ -0,0 +1,258 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Storage Decision - Executive Summary</title>
+<link rel="preconnect" href="https://fonts.googleapis.com">
+<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+<link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500;600&family=IBM+Plex+Sans:wght@400;500;600;700&display=swap" rel="stylesheet">
+<style>
+  *{box-sizing:border-box}
+  html,body{margin:0;padding:0}
+  body{font-family:"IBM Plex Sans",system-ui,sans-serif;color:#16181d;background:#eef0f2;-webkit-font-smoothing:antialiased}
+  .mono{font-family:"IBM Plex Mono",monospace}
+
+  header.top{display:flex;align-items:center;gap:14px;height:56px;padding:0 22px;background:#fff;border-bottom:1px solid #e2e5ea;position:sticky;top:0;z-index:5}
+  .logo{display:flex;align-items:center;justify-content:center;width:30px;height:30px;border-radius:7px;background:#16181d;color:#fff;font-family:"IBM Plex Mono",monospace;font-weight:600;font-size:12px;flex:none}
+  .htitle{display:flex;flex-direction:column;line-height:1.25;white-space:nowrap}
+  .htitle b{font-size:14.5px;font-weight:600;letter-spacing:-.01em}
+  .htitle span{font-size:11.5px;color:#838a95}
+  .hright{margin-left:auto;display:flex;align-items:center;gap:8px}
+  .chip{font-family:"IBM Plex Mono",monospace;font-size:11px;color:#838a95;background:#f3f5f7;border:1px solid #e6e8ec;padding:4px 10px;border-radius:6px;text-decoration:none}
+  a.chip:hover{border-color:#1f5fdb;color:#1f5fdb}
+  .chip-go{display:inline-flex;align-items:center;gap:6px;font-size:12px;color:#fff;background:#1f5fdb;padding:5px 12px;border-radius:6px;font-weight:600;text-decoration:none}
+  .chip-go:hover{background:#1a4fb8}
+
+  .page{max-width:980px;margin:0 auto;padding:30px 24px 70px}
+  .card{background:#fff;border:1px solid #e6e8ec;border-radius:12px;padding:22px 26px;margin-bottom:16px}
+  h1{margin:0 0 6px;font-size:24px;font-weight:700;letter-spacing:-.02em}
+  h2{margin:0 0 12px;font-size:16px;font-weight:600;letter-spacing:-.01em}
+  p{margin:0 0 10px;font-size:13.5px;color:#3d424b;line-height:1.65}
+  p:last-child{margin-bottom:0}
+  .muted{color:#838a95}
+  a.ext{color:#1f5fdb;text-decoration:none}
+  a.ext:hover{text-decoration:underline}
+  a.fact{color:inherit;text-decoration:underline dashed;text-decoration-thickness:1px;text-underline-offset:3px}
+  a.fact:hover{text-decoration-style:solid}
+
+  .hero{display:grid;grid-template-columns:1fr 1fr;gap:16px;margin-bottom:16px}
+  .hero .card{margin:0}
+  .hero .cap{font-family:"IBM Plex Mono",monospace;font-size:10.5px;letter-spacing:.1em;color:#a4abb4;margin-bottom:10px}
+  .hero .big{font-size:15.5px;line-height:1.6;color:#16181d}
+  .verdict{border:1px solid #bfe6cf;background:#e9f8ef;border-radius:12px;padding:16px 22px;display:flex;align-items:flex-start;gap:16px;margin-bottom:20px}
+  .verdict .mark{width:34px;height:34px;border-radius:50%;background:#18a05a;color:#fff;display:flex;align-items:center;justify-content:center;font-size:18px;font-weight:700;flex:none;margin-top:2px}
+  .verdict b{color:#0f7a45}
+  .verdict p{margin:0 0 8px;font-size:14.5px}
+  table.vtab{border-collapse:collapse;width:100%}
+  table.vtab th{text-align:left;font-size:11px;font-weight:600;padding:5px 16px 5px 0;border-bottom:1px solid #cdeadb}
+  table.vtab th.win{color:#0f7a45}
+  table.vtab th.lose{color:#b23b4e}
+  table.vtab td{font-size:13px;color:#2b3038;padding:7px 16px 7px 0;border-bottom:1px solid #d9f0e3;line-height:1.5;vertical-align:top}
+  table.vtab tr:last-child td{border-bottom:0}
+  table.vtab td.crit{font-weight:600;color:#16181d;width:26%}
+  table.vtab td.lose{color:#b23b4e}
+
+  .kpis{display:grid;grid-template-columns:repeat(3,1fr);gap:16px;margin-bottom:16px}
+  .kpi{background:#fff;border:1px solid #e6e8ec;border-radius:12px;padding:18px 20px}
+  .kpi .v{font-family:"IBM Plex Mono",monospace;font-size:34px;font-weight:600;line-height:1;color:#16181d}
+  .kpi .l{font-size:12.5px;color:#5b6472;margin-top:9px;line-height:1.5}
+
+  table.duel{width:100%;border-collapse:collapse}
+  table.duel th{padding:6px 14px;text-align:left;font-size:12px;border-bottom:2px solid #eef0f2;vertical-align:bottom}
+  table.duel th .pill{display:inline-flex;padding:3px 10px;border-radius:20px;font-size:10.5px;font-weight:600;margin-top:4px}
+  table.duel th.win{color:#0f7a45}
+  table.duel th.win .pill{color:#0f7a45;background:#e7f6ee}
+  table.duel th.lose{color:#b23b4e}
+  table.duel th.lose .pill{color:#b23b4e;background:#fbe9ec}
+  table.duel td{padding:7px 14px;font-size:13px;color:#3d424b;border-bottom:1px solid #f4f5f7;line-height:1.45;vertical-align:top}
+  table.duel td.crit{font-weight:600;color:#16181d;width:26%}
+  table.duel td.crit small{display:block;font-weight:400;color:#a4abb4;margin-top:2px}
+  table.duel td .ok{color:#0f7a45;font-weight:700;margin-right:6px}
+  table.duel td .no{color:#b23b4e;font-weight:700;margin-right:6px}
+  table.duel td.more{width:12%;text-align:right;white-space:nowrap}
+
+  .panels{display:grid;grid-template-columns:1fr 1fr;gap:22px;margin-top:6px}
+  .panel h3{margin:0 0 12px;font-size:12.5px;font-weight:600;color:#5b6472;text-transform:uppercase;letter-spacing:.04em}
+  .brow{display:flex;align-items:center;gap:12px;margin-bottom:12px}
+  .brow .lab{width:118px;flex:none;font-size:12.5px;color:#3d424b}
+  .brow .trk{flex:1;background:#f1f3f6;border-radius:5px;height:20px;overflow:hidden;position:relative}
+  .brow .fill{height:20px;border-radius:0 4px 4px 0;min-width:3px}
+  .brow .inval{position:absolute;top:50%;transform:translateY(-50%);font-family:"IBM Plex Mono",monospace;font-size:12px;font-weight:600;white-space:nowrap}
+  .brow .inval.in-track{left:12px;color:#16181d}
+  .brow .inval.in-fill{right:12px;color:#fff}
+  .brow:hover .trk{background:#eaedf1}
+  .fill.pg{background:#1f5fdb}
+  .fill.rdf{background:#c0476b}
+
+  .steps{counter-reset:st;list-style:none;margin:6px 0 0;padding:0}
+  .steps li{counter-increment:st;display:flex;gap:14px;align-items:flex-start;padding:10px 0;font-size:13.5px;color:#3d424b;line-height:1.6;border-bottom:1px solid #f4f5f7}
+  .steps li:last-child{border-bottom:0}
+  .steps li::before{content:counter(st);width:26px;height:26px;border-radius:50%;background:#eaf1fe;color:#1f5fdb;font-weight:700;font-size:13px;display:flex;align-items:center;justify-content:center;flex:none;margin-top:1px}
+
+  .foot{font-size:11.5px;color:#a4abb4;border-top:1px solid #e6e8ec;padding-top:16px;margin-top:26px;line-height:1.8}
+  @media(max-width:840px){.hero,.kpis,.panels{grid-template-columns:1fr}}
+</style>
+</head>
+<body>
+
+<header class="top">
+  <div class="logo">SE</div>
+  <div class="htitle">
+    <b>Storage Decision</b>
+    <span>Executive summary &middot; laboratory measurement data</span>
+  </div>
+  <div class="hright">
+    <a class="chip" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">Git repository</a>
+    <a class="chip-go" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html" target="_blank" rel="noopener">Open the full study &rarr;</a>
+  </div>
+</header>
+
+<div class="page">
+
+  <h1>Which database should hold the laboratory's data?</h1>
+
+  <div class="hero">
+    <div class="card">
+      <div class="cap">THE PROBLEM</div>
+      <p class="big"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#premises" target="_blank" rel="noopener"><b>600 GB</b></a> of laboratory data today.<br/><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#premises" target="_blank" rel="noopener"><b>6 TB</b></a> within a few years.<br/>It needs one working database.</p>
+    </div>
+    <div class="card">
+      <div class="cap">THE CANDIDATES</div>
+      <p class="big"><b>Relational</b> (PostgreSQL)<br/>vs<br/><b>Graph</b> (RDF triplestore).</p>
+    </div>
+  </div>
+
+  <div class="verdict">
+    <div class="mark">&#10003;</div>
+    <div style="flex:1">
+      <p><b>Decision: Relational vs Graph (RDF):</b></p>
+      <table class="vtab">
+        <tr>
+          <th></th>
+          <th class="win">Relational (PostgreSQL)</th>
+          <th class="lose">Graph (RDF)</th>
+        </tr>
+        <tr>
+          <td class="crit">Average report, today</td>
+          <td><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">~6 seconds</a></td>
+          <td class="lose"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">~1.6 hours</a></td>
+        </tr>
+        <tr>
+          <td class="crit">Average report at 6 TB</td>
+          <td><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">~30 seconds</a></td>
+          <td class="lose"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">~12 hours</a></td>
+        </tr>
+        <tr>
+          <td class="crit">Hardware at 6 TB</td>
+          <td><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#hardware" target="_blank" rel="noopener">one ordinary server</a></td>
+          <td class="lose"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#hardware" target="_blank" rel="noopener">a 14-server cluster</a></td>
+        </tr>
+        <tr>
+          <td class="crit">Ontology</td>
+          <td colspan="2">remains accessible with Relational (PostgreSQL)</td>
+        </tr>
+      </table>
+    </div>
+  </div>
+
+  <div class="kpis">
+    <div class="kpi"><div class="v"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#kpis" target="_blank" rel="noopener">1,000x</a></div><div class="l">faster on the average everyday report</div></div>
+    <div class="kpi"><div class="v"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#kpis" target="_blank" rel="noopener">$494k saved</a></div><div class="l">hardware at 6 TB: $33.6k for PGSQL server instead of a $527.7k RDF cluster</div></div>
+    <div class="kpi"><div class="v"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#kpis" target="_blank" rel="noopener">1 vs 14</a></div><div class="l">servers needed at 6 TB</div></div>
+  </div>
+
+  <div class="card">
+    <h2>Side by side</h2>
+    <table class="duel">
+      <tr>
+        <th></th>
+        <th class="win">Relational (PostgreSQL)<br><span class="pill">&#10003; recommended</span></th>
+        <th class="lose">Graph (RDF triplestore)<br><span class="pill">&#10007; rejected at scale</span></th>
+        <th></th>
+      </tr>
+      <tr>
+        <td class="crit">Average report, today<small>at 600 GB</small></td>
+        <td><span class="ok">&#10003;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">about 6 seconds</a></td>
+        <td><span class="no">&#10007;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">about 1.6 hours; 3 of 7 tasks need more than a day</a></td>
+        <td class="more"><a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html#proj" target="_blank" rel="noopener">details &rarr;</a></td>
+      </tr>
+      <tr>
+        <td class="crit">Average report at 6 TB<small>the 3-year horizon</small></td>
+        <td><span class="ok">&#10003;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">about 30 seconds</a></td>
+        <td><span class="no">&#10007;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">about 12 hours; 5 of 7 tasks need a day or more</a></td>
+        <td class="more"><a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html#proj" target="_blank" rel="noopener">details &rarr;</a></td>
+      </tr>
+      <tr>
+        <td class="crit">Servers needed today<small>at 600 GB</small></td>
+        <td><span class="ok">&#10003;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#hardware" target="_blank" rel="noopener">one standard server, $10.5k</a></td>
+        <td><span class="no">&#10007;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#hardware" target="_blank" rel="noopener">already two servers with 1.4 TB of memory, $58.7k</a></td>
+        <td class="more"><a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html#hw" target="_blank" rel="noopener">details &rarr;</a></td>
+      </tr>
+      <tr>
+        <td class="crit">Servers at 6 TB</td>
+        <td><span class="ok">&#10003;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#hardware" target="_blank" rel="noopener">still one server, $33.6k</a></td>
+        <td><span class="no">&#10007;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#hardware" target="_blank" rel="noopener">a 14-server cluster, $527.7k</a></td>
+        <td class="more"><a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html#hw" target="_blank" rel="noopener">details &rarr;</a></td>
+      </tr>
+      <tr>
+        <td class="crit">Disk for the same data<small>at 6 TB</small></td>
+        <td><span class="ok">&#10003;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#storage" target="_blank" rel="noopener">21 TB</a></td>
+        <td><span class="no">&#10007;</span><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#storage" target="_blank" rel="noopener">91 TB - the graph form stores the same data 4.4x bigger</a></td>
+        <td class="more"><a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html#storage" target="_blank" rel="noopener">details &rarr;</a></td>
+      </tr>
+    </table>
+  </div>
+
+  <div class="card">
+    <h2>Average everyday report</h2>
+    <div class="panels">
+      <div class="panel">
+        <h3>Today &middot; 600 GB</h3>
+        <div class="brow" title="PostgreSQL, average of the 7 everyday tasks at 600 GB: about 6 seconds (projected)">
+          <div class="lab">PostgreSQL</div>
+          <div class="trk"><div class="fill pg" style="width:0.4%"></div><span class="inval in-track"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">6 s</a></span></div>
+        </div>
+        <div class="brow" title="RDF triplestore, average of the 7 everyday tasks at 600 GB: about 1.6 hours (projected)">
+          <div class="lab">Graph (RDF)</div>
+          <div class="trk"><div class="fill rdf" style="width:100%"></div><span class="inval in-fill"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">1.6 h</a></span></div>
+        </div>
+      </div>
+      <div class="panel">
+        <h3>In a few years &middot; 6 TB</h3>
+        <div class="brow" title="PostgreSQL, average of the 7 everyday tasks at 6 TB: about 30 seconds (projected)">
+          <div class="lab">PostgreSQL</div>
+          <div class="trk"><div class="fill pg" style="width:0.4%"></div><span class="inval in-track"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">30 s</a></span></div>
+        </div>
+        <div class="brow" title="RDF triplestore, average of the 7 everyday tasks at 6 TB: about 12 hours (projected)">
+          <div class="lab">Graph (RDF)</div>
+          <div class="trk"><div class="fill rdf" style="width:100%"></div><span class="inval in-fill"><a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#performance" target="_blank" rel="noopener">12 h</a></span></div>
+        </div>
+      </div>
+    </div>
+    <p class="muted" style="margin-top:14px">Bars are to scale within each panel - the blue bar is real, it is just that small. An everyday report that takes hours is a report nobody runs.</p>
+  </div>
+
+  <div class="card">
+    <h2>What about the other formats?</h2>
+    <p><b>CSV</b> stays exactly where it belongs - the raw archive coming off the instruments (best compression, byte-reproducible). <b>JSON-LD</b> stays the exchange format for sending data between laboratories. Neither is a working database, so neither was a candidate. <b>SQLite</b> is excellent for a single user <a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#notes" target="_blank" rel="noopener">up to ~600 GB</a> but does not scale to the shared 6 TB horizon.</p>
+    <p><b>And the ontology?</b> It survives intact: semantic, ontology-style access is provided as a <b>virtual layer on top of PostgreSQL</b> (e.g. Ontop OBDA) - the graph view without the graph database's cost.</p>
+  </div>
+
+  <div class="card">
+    <h2>Recommended next steps</h2>
+    <ol class="steps">
+      <li>Adopt <b>PostgreSQL as the system of record</b> for laboratory measurement data; keep the raw CSV archive as the source of truth for reprocessing.</li>
+      <li>Provide ontology and semantic queries as a <b>virtual RDF layer</b> over PostgreSQL - no data is copied into a triplestore.</li>
+      <li>Revisit hardware once volume approaches 6 TB: the measured plan is <a class="fact" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html#hardware" target="_blank" rel="noopener">one 16-core server with 551 GB RAM (~$33.6k at July 2026 prices)</a>.</li>
+    </ol>
+  </div>
+
+  <div class="foot">
+    Figures at 600 GB and 6 TB are model projections calibrated on a measured benchmark (7 everyday retrieval tasks x 5 storage formats, every result checksum-validated); the "average report" is the geometric mean of the seven tasks, the same averaging the study's scoring uses. Method, raw data and hardware price sources: <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision_Facts.html" target="_blank" rel="noopener">facts ledger (every number traced to its artifact)</a> &middot; <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html" target="_blank" rel="noopener">full interactive study</a> &middot; <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">git.boskadoff.com/Public/LabDataStorageEvaluation</a>.<br>
+    Study: Mary Goncharenko (Tribology Laboratory, University of Florida; internship at Sandia National Laboratories), with the technical assistance of Vasiliy Goncharenko (SoftCreator, LLC).
+  </div>
+
+</div>
+</body>
+</html>

docs/research/Tribology_Storage_Decision_Facts.html

@@ -0,0 +1,131 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Decision Facts Ledger - Storage Decision</title>
+<link rel="preconnect" href="https://fonts.googleapis.com">
+<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+<link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500;600&family=IBM+Plex+Sans:wght@400;500;600;700&display=swap" rel="stylesheet">
+<style>
+  *{box-sizing:border-box}
+  html,body{margin:0;padding:0}
+  body{font-family:"IBM Plex Sans",system-ui,sans-serif;color:#16181d;background:#eef0f2;-webkit-font-smoothing:antialiased}
+  .mono{font-family:"IBM Plex Mono",monospace}
+  header.top{display:flex;align-items:center;gap:14px;height:56px;padding:0 22px;background:#fff;border-bottom:1px solid #e2e5ea;position:sticky;top:0;z-index:5}
+  .logo{display:flex;align-items:center;justify-content:center;width:30px;height:30px;border-radius:7px;background:#16181d;color:#fff;font-family:"IBM Plex Mono",monospace;font-weight:600;font-size:12px;flex:none}
+  .htitle{display:flex;flex-direction:column;line-height:1.25;white-space:nowrap}
+  .htitle b{font-size:14.5px;font-weight:600;letter-spacing:-.01em}
+  .htitle span{font-size:11.5px;color:#838a95}
+  .hright{margin-left:auto;display:flex;align-items:center;gap:8px}
+  .chip{font-family:"IBM Plex Mono",monospace;font-size:11px;color:#838a95;background:#f3f5f7;border:1px solid #e6e8ec;padding:4px 10px;border-radius:6px;text-decoration:none}
+  a.chip:hover{border-color:#1f5fdb;color:#1f5fdb}
+  .page{max-width:1080px;margin:0 auto;padding:28px 24px 70px}
+  h1{margin:0 0 6px;font-size:22px;font-weight:700;letter-spacing:-.02em}
+  h2{margin:26px 0 10px;font-size:15px;font-weight:600}
+  p{margin:0 0 10px;font-size:13px;color:#3d424b;line-height:1.6;max-width:940px}
+  .muted{color:#838a95}
+  a{color:#1f5fdb;text-decoration:none}
+  a:hover{text-decoration:underline}
+  .card{background:#fff;border:1px solid #e6e8ec;border-radius:11px;padding:6px 18px 10px;margin-bottom:8px;overflow-x:auto}
+  table{width:100%;border-collapse:collapse;min-width:760px}
+  th{font-size:10.5px;text-transform:uppercase;letter-spacing:.04em;color:#838a95;font-weight:600;text-align:left;padding:9px 12px;border-bottom:1px solid #eef0f2}
+  td{font-size:12.5px;color:#3d424b;padding:9px 12px;border-bottom:1px solid #f4f5f7;vertical-align:top;line-height:1.55}
+  tr:last-child td{border-bottom:0}
+  td.id{font-family:"IBM Plex Mono",monospace;font-weight:600;color:#16181d;white-space:nowrap}
+  td.val{font-family:"IBM Plex Mono",monospace;white-space:nowrap}
+  .foot{font-size:11.5px;color:#a4abb4;border-top:1px solid #e6e8ec;padding-top:14px;margin-top:26px;line-height:1.7}
+</style>
+</head>
+<body>
+
+<header class="top">
+  <div class="logo">SE</div>
+  <div class="htitle">
+    <b>Decision Facts Ledger</b>
+    <span>Every fact on the Storage Decision page &middot; value, formula, source</span>
+  </div>
+  <div class="hright">
+    <a class="chip" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision.html" target="_blank" rel="noopener">&larr; Storage Decision</a>
+    <a class="chip" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html" target="_blank" rel="noopener">Full study</a>
+    <a class="chip" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">Git repository</a>
+  </div>
+</header>
+
+<div class="page">
+
+  <h1>Decision Facts Ledger</h1>
+  <p>Every number shown on the <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision.html" target="_blank" rel="noopener">Storage Decision</a> executive summary, traced to the artifact that proves it. All source files are committed in the public repository and regenerate deterministically from seed 20260711. Projection tables carry one row per format x task x scale; "geometric mean" is the same averaging the study's scoring uses.</p>
+  <p class="muted">Sources: <span class="mono">t6</span> = <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6_projected_all_scales.csv</a> &middot; <span class="mono">t7</span> = <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7_hardware_all_scales.csv</a> &middot; <span class="mono">t1</span> = <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t1_storage_footprint.csv" target="_blank" rel="noopener">t1_storage_footprint.csv</a> &middot; <span class="mono">prices</span> = <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/config/hw_prices.yaml" target="_blank" rel="noopener">hw_prices.yaml</a> &middot; <span class="mono">spec</span> = <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/docs/specs/10_extrapolation_model.md" target="_blank" rel="noopener">10_extrapolation_model.md</a></p>
+
+  <h2 id="premises">Premises</h2>
+  <div class="card">
+    <table>
+      <tr><th>ID</th><th>Fact on the page</th><th>Exact value</th><th>How obtained</th><th>Source</th></tr>
+      <tr><td class="id">P1</td><td>"600 GB of laboratory data today"</td><td class="val">600 GB</td><td>The laboratory's current operational volume; also the first extrapolation scale of the study (scale_gb = 600).</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/docs/specs/10_extrapolation_model.md" target="_blank" rel="noopener">spec</a>, <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">P2</td><td>"6 TB within a few years"</td><td class="val">6,000 GB</td><td>The largest extrapolation scale of the study (scale_gb = 6000), the growth horizon.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/docs/specs/10_extrapolation_model.md" target="_blank" rel="noopener">spec</a>, <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+    </table>
+  </div>
+
+  <h2 id="performance">Performance (average everyday report)</h2>
+  <div class="card">
+    <table>
+      <tr><th>ID</th><th>Fact on the page</th><th>Exact value</th><th>How obtained</th><th>Source</th></tr>
+      <tr><td class="id">F1</td><td>PostgreSQL, average report today: "~6 seconds"</td><td class="val">5.66 s</td><td>Geometric mean of the 7 projected_wall_s values where format = PostgreSQL, scale_gb = 600.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">F2</td><td>PostgreSQL, average report at 6 TB: "~30 seconds"</td><td class="val">29.6 s</td><td>Geometric mean, format = PostgreSQL, scale_gb = 6000.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">F3</td><td>Graph (RDF), average report today: "~1.6 hours"</td><td class="val">5,902 s = 1.64 h</td><td>Geometric mean, format = RDF triplestore, scale_gb = 600.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">F4</td><td>Graph (RDF), average report at 6 TB: "~12 hours"</td><td class="val">43,157 s = 11.99 h</td><td>Geometric mean, format = RDF triplestore, scale_gb = 6000.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">F5</td><td>"3 of 7 tasks need more than a day" (RDF, today)</td><td class="val">q3 2.1 d &middot; q5 2.8 d &middot; q7 1.9 d</td><td>Count of RDF rows at scale_gb = 600 with projected_wall_s &gt; 86,400.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">F6</td><td>"5 of 7 tasks need a day or more" (RDF, 6 TB)</td><td class="val">q2, q3, q4, q5, q7</td><td>Count of RDF rows at scale_gb = 6000 with projected_wall_s &gt;= 86,400.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+    </table>
+  </div>
+
+  <h2 id="hardware">Hardware and cost</h2>
+  <div class="card">
+    <table>
+      <tr><th>ID</th><th>Fact on the page</th><th>Exact value</th><th>How obtained</th><th>Source</th></tr>
+      <tr><td class="id">F7</td><td>"one standard server, $10.5k" (PostgreSQL, today)</td><td class="val">$10,500 &middot; 1 node</td><td>Row format = PostgreSQL, scale_gb = 600: est_cost_usd, nodes; bill of materials in the configuration column.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+      <tr><td class="id">F8</td><td>"already two servers with 1.4 TB of memory, $58.7k" (RDF, today)</td><td class="val">$58,740 &middot; 2 nodes &middot; 1,404.1 GB RAM</td><td>Row format = RDF triplestore, scale_gb = 600: est_cost_usd, nodes, ram_gb.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+      <tr><td class="id">F9</td><td>"still one server, $33.6k" / "one 16-core server with 551 GB RAM" (PostgreSQL, 6 TB)</td><td class="val">$33,586 &middot; 1 node &middot; 16 cores &middot; 551.3 GB</td><td>Row format = PostgreSQL, scale_gb = 6000: est_cost_usd, nodes, cores, ram_gb.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+      <tr><td class="id">F10</td><td>"a 14-server cluster, $527.7k" (RDF, 6 TB)</td><td class="val">$527,732 &middot; 14 nodes</td><td>Row format = RDF triplestore, scale_gb = 6000: est_cost_usd, nodes.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+      <tr><td class="id">F11</td><td>"July 2026 prices"</td><td class="val">as_of 2026-07-11</td><td>Street prices with per-component source URLs recorded in the operator-editable price file.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/config/hw_prices.yaml" target="_blank" rel="noopener">prices</a></td></tr>
+    </table>
+  </div>
+
+  <h2 id="storage">Storage</h2>
+  <div class="card">
+    <table>
+      <tr><th>ID</th><th>Fact on the page</th><th>Exact value</th><th>How obtained</th><th>Source</th></tr>
+      <tr><td class="id">F12</td><td>"21 TB" disk (PostgreSQL, 6 TB)</td><td class="val">20.855 TB</td><td>Row format = PostgreSQL, scale_gb = 6000: disk_tb (includes 30% free-space headroom).</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+      <tr><td class="id">F13</td><td>"91 TB" disk (RDF, 6 TB)</td><td class="val">91.268 TB</td><td>Row format = RDF triplestore, scale_gb = 6000: disk_tb.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+      <tr><td class="id">F14</td><td>"stores the same data 4.4x bigger"</td><td class="val">11.70 / 2.67 = 4.38</td><td>Ratio of the measured working-footprint coefficients vs raw CSV (RDF 11.70x over PostgreSQL 2.67x).</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t1_storage_footprint.csv" target="_blank" rel="noopener">t1</a></td></tr>
+    </table>
+  </div>
+
+  <h2 id="kpis">Headline figures</h2>
+  <div class="card">
+    <table>
+      <tr><th>ID</th><th>Fact on the page</th><th>Exact value</th><th>How obtained</th><th>Source</th></tr>
+      <tr><td class="id">K1</td><td>"1,000x faster on the average everyday report"</td><td class="val">1,043x today &middot; 1,459x at 6 TB</td><td>F3 / F1 = 5,902 / 5.66; F4 / F2 = 43,157 / 29.6. Stated conservatively as 1,000x.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">K2</td><td>"$494k saved" (hardware at 6 TB)</td><td class="val">$494,146</td><td>F10 - F9 = 527,732 - 33,586.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+      <tr><td class="id">K3</td><td>"1 vs 14 servers needed at 6 TB"</td><td class="val">nodes: 1 vs 14</td><td>nodes column, rows PostgreSQL/6000 and RDF triplestore/6000.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t7_hardware_all_scales.csv" target="_blank" rel="noopener">t7</a></td></tr>
+    </table>
+  </div>
+
+  <h2 id="notes">Notes</h2>
+  <div class="card">
+    <table>
+      <tr><th>ID</th><th>Fact on the page</th><th>Exact value</th><th>How obtained</th><th>Source</th></tr>
+      <tr><td class="id">N1</td><td>"SQLite is excellent for a single user up to ~600 GB"</td><td class="val">all 7 tasks OK at 600 GB; q5 = 1.6 h, q7 = 1.5 h at 1.2 TB</td><td>SQLite rows in t6: every flag OK at scale_gb = 600; its single-threaded full scans exceed the 1-hour bound from scale_gb = 1200.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/src/branch/master/out/report/tables/t6_projected_all_scales.csv" target="_blank" rel="noopener">t6</a></td></tr>
+      <tr><td class="id">N2</td><td>Method behind every projected number</td><td class="val">measured 35-cell benchmark</td><td>Scaling laws calibrated on the measured corpus; every measured run checksum-validated. Full methodology and measured medians in the interactive study.</td><td><a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Evaluation.html" target="_blank" rel="noopener">full study</a></td></tr>
+    </table>
+  </div>
+
+  <div class="foot">
+    Part of the LabDataStorageEvaluation study &middot; run of 2026-07-11, seed 20260711 &middot; canonical source: <a href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">git.boskadoff.com/Public/LabDataStorageEvaluation</a>.<br>
+    Study: Mary Goncharenko (Tribology Laboratory, University of Florida; internship at Sandia National Laboratories), with the technical assistance of Vasiliy Goncharenko (SoftCreator, LLC).
+  </div>
+
+</div>
+</body>
+</html>

docs/research/Tribology_Storage_Evaluation.html

@@ -0,0 +1,814 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Storage Format Evaluation - Tribology Lab Data</title>
+<link rel="preconnect" href="https://fonts.googleapis.com">
+<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+<link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500;600&family=IBM+Plex+Sans:wght@400;500;600;700&display=swap" rel="stylesheet">
+<style>
+  *{box-sizing:border-box}
+  html,body{margin:0;padding:0}
+  body{font-family:"IBM Plex Sans",system-ui,sans-serif;color:#16181d;background:#eef0f2;-webkit-font-smoothing:antialiased}
+  ::-webkit-scrollbar{width:10px;height:10px}
+  ::-webkit-scrollbar-thumb{background:#cfd4da;border-radius:6px;border:2px solid #eef0f2}
+  ::-webkit-scrollbar-track{background:transparent}
+  .mono{font-family:"IBM Plex Mono",monospace}
+
+  .app{display:flex;flex-direction:column;height:100vh;width:100%;overflow:hidden;background:#eef0f2}
+  header.top{display:flex;align-items:center;gap:14px;height:56px;flex:none;padding:0 22px;background:#fff;border-bottom:1px solid #e2e5ea}
+  .logo{display:flex;align-items:center;justify-content:center;width:30px;height:30px;border-radius:7px;background:#16181d;color:#fff;font-family:"IBM Plex Mono",monospace;font-weight:600;font-size:12px}
+  .htitle{display:flex;flex-direction:column;line-height:1.25;white-space:nowrap}
+  .htitle b{font-size:14.5px;font-weight:600;letter-spacing:-.01em}
+  .htitle span{font-size:11.5px;color:#838a95}
+  .hright{margin-left:auto;display:flex;align-items:center;gap:8px}
+  .chip{font-family:"IBM Plex Mono",monospace;font-size:11px;color:#838a95;background:#f3f5f7;border:1px solid #e6e8ec;padding:4px 10px;border-radius:6px}
+  .chip-ok{display:inline-flex;align-items:center;gap:6px;font-size:12px;color:#0f7a45;background:#e7f6ee;padding:4px 10px;border-radius:6px;font-weight:600}
+  .chip-ok i{width:7px;height:7px;border-radius:7px;background:#18a05a}
+
+  .frame{display:flex;flex:1;min-height:0}
+  nav.side{width:252px;flex:none;background:#fff;border-right:1px solid #e2e5ea;padding:14px 12px;display:flex;flex-direction:column;overflow-y:auto}
+  .navcap{font-family:"IBM Plex Mono",monospace;font-size:10px;letter-spacing:.1em;color:#a4abb4;padding:6px 11px 8px}
+  nav.side a{display:flex;align-items:center;gap:11px;width:100%;padding:8px 11px;border-radius:8px;cursor:pointer;text-decoration:none;font-size:13.5px;margin:1px 0;color:#475059;font-weight:500}
+  nav.side a .tag{width:24px;height:19px;border-radius:5px;display:flex;align-items:center;justify-content:center;font-family:"IBM Plex Mono",monospace;font-size:9px;font-weight:600;flex:none;background:#eceef1;color:#8a909b}
+  nav.side a.active{background:#eaf1fe;color:#16181d;font-weight:600}
+  nav.side a.active .tag{background:#1f5fdb;color:#fff}
+  .sidefoot{margin-top:auto;padding:14px 11px 6px;border-top:1px solid #eef0f2}
+  .sidefoot .cap{font-family:"IBM Plex Mono",monospace;font-size:10px;letter-spacing:.08em;color:#a4abb4;margin-bottom:8px}
+  .sidefoot .txt{font-size:12px;color:#6b727c;line-height:1.6}
+
+  main{flex:1;min-width:0;overflow-y:auto;padding:26px 30px 60px}
+  main section{display:none}
+  main section.on{display:block}
+  h1{margin:0;font-size:21px;font-weight:600;letter-spacing:-.02em}
+  .sub{margin:4px 0 18px;font-size:13px;color:#838a95;max-width:900px;line-height:1.55}
+  .card{background:#fff;border:1px solid #e6e8ec;border-radius:11px;padding:18px 20px;margin-bottom:14px}
+  .card h3{margin:0 0 10px;font-size:14px;font-weight:600}
+  .card p{margin:0 0 10px;font-size:13px;color:#3d424b;line-height:1.6;max-width:900px}
+  .card p:last-child{margin-bottom:0}
+  .card ul{margin:6px 0 4px;padding-left:20px}
+  .card li{font-size:13px;color:#3d424b;line-height:1.65;margin-bottom:6px;max-width:860px}
+  .muted{color:#838a95}
+  .kpis{display:grid;grid-template-columns:repeat(5,1fr);gap:14px;margin-bottom:16px}
+  .kpis.four{grid-template-columns:repeat(4,1fr)}
+  .kpi{background:#fff;border:1px solid #e6e8ec;border-radius:10px;padding:15px 16px}
+  .kpi .l{font-size:11px;color:#838a95;font-weight:500;text-transform:uppercase;letter-spacing:.04em}
+  .kpi .v{font-family:"IBM Plex Mono",monospace;font-size:24px;font-weight:600;margin-top:7px;line-height:1}
+  .kpi .s{font-size:11.5px;color:#a4abb4;margin-top:5px}
+  .v-bad{color:#b23b4e}.v-good{color:#0f7a45}.v-neutral{color:#16181d}
+
+  table.data{width:100%;border-collapse:collapse}
+  table.data th{font-size:10.5px;text-transform:uppercase;letter-spacing:.04em;color:#838a95;font-weight:600;text-align:left;padding:9px 12px;background:#f7f8fa;border-bottom:1px solid #eef0f2}
+  table.data td{font-size:12.5px;color:#3d424b;padding:10px 12px;border-bottom:1px solid #f4f5f7;vertical-align:top}
+  table.data th.num, table.data td.num{text-align:right;font-family:"IBM Plex Mono",monospace}
+  table.data td.mono{font-family:"IBM Plex Mono",monospace}
+
+  .hbars{display:flex;flex-direction:column;gap:11px;margin-top:6px}
+  .hbar{display:flex;align-items:center;gap:14px}
+  .hbar .lab{width:190px;flex:none;font-size:12.5px;color:#3d424b}
+  .hbar .lab small{display:block;color:#a4abb4;font-size:11px}
+  .hbar .trk{flex:1;background:#f1f3f6;border-radius:6px;overflow:hidden}
+  .hbar .fill{height:14px;border-radius:6px;min-width:2px}
+  .hbar .val{font-family:"IBM Plex Mono",monospace;font-size:12.5px;color:#16181d;width:130px;flex:none;text-align:right}
+  .legend{display:flex;gap:16px;flex-wrap:wrap;margin:2px 0 10px}
+  .legend span{display:inline-flex;align-items:center;gap:6px;font-size:11.5px;color:#5b6472}
+  .legend i{width:10px;height:10px;border-radius:3px;display:inline-block}
+
+  .stack{display:flex;align-items:center;gap:14px;margin-bottom:11px}
+  .stack .lab{width:190px;flex:none;font-size:12.5px;color:#3d424b}
+  .stack .trk{flex:1;display:flex;background:#f1f3f6;border-radius:6px;overflow:hidden;height:16px}
+  .stack .seg{height:16px}
+  .stack .val{font-family:"IBM Plex Mono",monospace;font-size:12.5px;width:70px;flex:none;text-align:right;font-weight:600}
+
+  .tabs{display:inline-flex;gap:4px;background:#eceef1;border-radius:8px;padding:3px;margin:0 0 14px}
+  .tabs button{border:0;background:transparent;font-family:"IBM Plex Sans",system-ui,sans-serif;font-size:12px;font-weight:600;color:#5b6472;padding:5px 12px;border-radius:6px;cursor:pointer}
+  .tabs button.on{background:#fff;color:#16181d;box-shadow:0 1px 3px rgba(20,30,50,.08)}
+
+  .callout{border-left:3px solid #1f5fdb;background:#f5f8fe;border-radius:0 9px 9px 0;padding:12px 16px;font-size:13px;color:#2b3038;line-height:1.6;margin:12px 0;max-width:900px}
+  .callout.warn{border-left-color:#b23b4e;background:#fdf3f5}
+  .callout.good{border-left-color:#18a05a;background:#f0faf4}
+  .pill{display:inline-flex;align-items:center;padding:2px 9px;border-radius:20px;font-size:10.5px;font-weight:600;white-space:nowrap}
+  .pill.ok{color:#0f7a45;background:#e7f6ee}
+  .pill.imp{color:#a06a12;background:#fdf3e0}
+  .pill.fail{color:#b23b4e;background:#fbe9ec}
+  .pill.rec{color:#0f7a45;background:#e7f6ee}
+  .pill.base{color:#5b6472;background:#eef0f2}
+
+  pre.env{background:#16181d;color:#dfe3e9;border-radius:9px;padding:14px 16px;font-family:"IBM Plex Mono",monospace;font-size:11.5px;line-height:1.65;overflow-x:auto;margin:10px 0 0}
+  .tree{font-family:"IBM Plex Mono",monospace;font-size:12px;line-height:1.8;color:#3d424b;background:#f7f8fa;border:1px solid #eef0f2;border-radius:9px;padding:14px 18px;margin-top:8px}
+  .grid2{display:grid;grid-template-columns:1fr 1fr;gap:14px}
+  @media(max-width:1100px){.grid2{grid-template-columns:1fr}.kpis{grid-template-columns:repeat(2,1fr)}}
+  .foot{font-size:11.5px;color:#a4abb4;border-top:1px solid #e6e8ec;padding-top:14px;margin-top:30px;line-height:1.7}
+  .src a, a.ext{color:#1f5fdb;text-decoration:none;word-break:break-all}
+  .src a:hover, a.ext:hover{text-decoration:underline}
+  a.chip{text-decoration:none}
+  a.chip:hover{border-color:#1f5fdb;color:#1f5fdb}
+  svg text{font-family:"IBM Plex Sans",system-ui,sans-serif}
+</style>
+</head>
+<body>
+<div class="app">
+
+  <header class="top">
+    <div class="logo">SE</div>
+    <div class="htitle">
+      <b>Storage Format Evaluation</b>
+      <span>Tribology lab data &middot; CSV vs JSON-LD vs SQLite vs PostgreSQL vs RDF</span>
+    </div>
+    <div class="hright">
+      <a class="chip" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">Git repository</a>
+      <span class="chip">7 queries x 5 formats &middot; 35 cells</span>
+      <span class="chip">seed 20260711</span>
+      <span class="chip-ok"><i></i>Measured 2026-07-11 &middot; all checksums OK</span>
+    </div>
+  </header>
+
+  <div class="frame">
+
+    <nav class="side" id="sidenav">
+      <div class="navcap">NAVIGATE</div>
+      <a href="#dashboard" class="active"><span class="tag">DB</span><span>Dashboard</span></a>
+      <a href="#lab"><span class="tag">LM</span><span>Lab Model &amp; Corpus</span></a>
+      <a href="#method"><span class="tag">ME</span><span>Methodology</span></a>
+      <a href="#storage"><span class="tag">SF</span><span>Storage Footprint</span></a>
+      <a href="#measured"><span class="tag">MR</span><span>Measured Results</span></a>
+      <a href="#proj"><span class="tag">PR</span><span>Projections 0.6-6 TB</span></a>
+      <a href="#hw"><span class="tag">HW</span><span>Hardware &amp; Cost</span></a>
+      <a href="#score"><span class="tag">SC</span><span>Scoring &amp; Verdict</span></a>
+      <a href="#about"><span class="tag">AB</span><span>About</span></a>
+      <div class="sidefoot">
+        <div class="cap">CONTEXT</div>
+        <div class="txt">Final results of the LabDataStorageEvaluation pipeline (tasks 01-11): a simulated Pt-Au tribology laboratory corpus benchmarked across five storage formats. Sources of truth: out/report/ and out/bench/.</div>
+        <div class="cap" style="margin-top:12px">AUTHOR &middot; SOURCE</div>
+        <div class="txt">Mary Goncharenko - University of Florida / Sandia National Laboratories.<br>
+        <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">git.boskadoff.com/Public/LabDataStorageEvaluation</a></div>
+      </div>
+    </nav>
+
+    <main id="main">
+
+      <!-- ================= DASHBOARD ================= -->
+      <section id="dashboard" class="on">
+        <h1>Dashboard</h1>
+        <p class="sub">A 141.8 MiB physically plausible CSV corpus (the canonical raw format) was converted into four other storage formats and benchmarked on seven laboratory retrieval scenarios with subprocess-isolated resource metering; results were extrapolated to 600 GB, 1.2 TB and 6 TB. Every measured run validated its rows against canonical results (PostgreSQL cross-validated with SQLite, tolerance 1e-9). Short on time? The one-page management summary is <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision.html" target="_blank" rel="noopener">Storage Decision</a>.</p>
+        <div class="kpis">
+          <div class="kpi"><div class="l">Benchmark cells</div><div class="v v-good">35 / 35</div><div class="s">OK; 0 invalid, 0 timeouts</div></div>
+          <div class="kpi"><div class="l">Corpus</div><div class="v v-neutral">141.8 MiB</div><div class="s">8,718 CSV files, byte-reproducible</div></div>
+          <div class="kpi"><div class="l">Winner @600 GB</div><div class="v v-good">58.6 / 80</div><div class="s">PostgreSQL weighted score</div></div>
+          <div class="kpi"><div class="l">Worst blow-up</div><div class="v v-bad">11.7x</div><div class="s">RDF store vs raw CSV size</div></div>
+          <div class="kpi"><div class="l">Session</div><div class="v v-neutral">916 s</div><div class="s">whole matrix, quiet machine</div></div>
+        </div>
+        <div class="callout good"><b>Recommendation (confirmed by data):</b> PostgreSQL as the system of record; CSV retained as the raw archive (best compressor, ~2.8x); hybrid JSON-LD for inter-lab exchange; RDF only as a virtual layer (e.g. Ontop OBDA) over PostgreSQL - a materialized triplestore fails at scale on speed, RAM and cost simultaneously.</div>
+        <div class="card">
+          <h3>Typical query time per format (geometric mean Q1-Q7)</h3>
+          <div id="dash-geo" class="hbars"></div>
+          <p class="muted" style="margin-top:12px">Left value: measured on the 150 MB corpus. Right value in parentheses: projected at 600 GB. Linear bars scaled within the measured values; every projected number in this document is labeled as projected.</p>
+        </div>
+        <div class="card">
+          <h3>Where each format breaks</h3>
+          <table class="data">
+            <thead><tr><th>Format</th><th>Strength</th><th>Breaking point</th><th>@600 GB flags</th></tr></thead>
+            <tbody id="dash-breaks"></tbody>
+          </table>
+        </div>
+      </section>
+
+      <!-- ================= LAB MODEL ================= -->
+      <section id="lab">
+        <h1>Lab Model &amp; Corpus</h1>
+        <p class="sub">A simulated tribology laboratory (Pt-Au LDRD study context, Sandia National Laboratories): Ti-6Al-4V coupons with a sputtered Cr adhesion layer and a Pt-Au composition-gradient coating, tested on a 6-probe tribometer in two environments. All data is generated deterministically from seed 20260711; regeneration is byte-identical (MANIFEST.csv sha256 proof).</p>
+        <div class="kpis four">
+          <div class="kpi"><div class="l">Deposition</div><div class="v v-neutral">4 &middot; 12 &middot; 588</div><div class="s">batches &middot; wafers &middot; coupons (7x7 grid)</div></div>
+          <div class="kpi"><div class="l">Friction testing</div><div class="v v-neutral">1,440</div><div class="s">tracks: 480 coupons x 3 replicates</div></div>
+          <div class="kpi"><div class="l">Bulk rows</div><div class="v v-neutral">4.32 M</div><div class="s">1.44M COF cycles + 2.88M loop points</div></div>
+          <div class="kpi"><div class="l">Characterization</div><div class="v v-neutral">235,200</div><div class="s">XRF points + nanoindentation, AFM, profilometry</div></div>
+        </div>
+        <div class="grid2">
+          <div class="card">
+            <h3>Data hierarchy</h3>
+            <div class="tree">batch (B721-B724, deposition matrix)
+&nbsp;&nbsp;-&gt; wafer (x3 per batch)
+&nbsp;&nbsp;&nbsp;&nbsp;-&gt; coupon (x49 per wafer; 480 friction + 108 reserve)
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;-&gt; xrf_map (400 pts) &middot; nanoindentation (25) &middot; afm &middot; profilometry (100)
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;-&gt; track (x3, counterface rotation)
+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;-&gt; cof_vs_cycle (1,000 rows) &middot; friction_loops (10x200) &middot; wear</div>
+            <p style="margin-top:10px">Runs R1-R4 map one-to-one onto batches; two environments (lab air / dry N2). The Q1 benchmark target track is <span class="mono">B722-W2-C13-T2</span>.</p>
+          </div>
+          <div class="card">
+            <h3>Physical models behind the numbers</h3>
+            <ul>
+              <li>COF: exponential run-in to a steady state depending on Au wt% and environment; noise floor 0.12.</li>
+              <li>Hardness / modulus: linear in Au wt% with measurement noise (25 indents per coupon).</li>
+              <li>Wear: Archard model, k0 = 1.5e-7 (lab air) / 6e-8 (dry N2) mm3/(N*m), log-normal noise.</li>
+              <li>Au gradient across wafer position; batch-dependent center 8 / 25 / 10 / 60 wt%.</li>
+            </ul>
+            <p>Benchmark queries therefore return physically meaningful answers (e.g. Q2 selects the Au 10 +/- 0.5 wt% composition window), not random noise.</p>
+          </div>
+        </div>
+        <div class="card">
+          <h3>The five storage formats under test</h3>
+          <table class="data">
+            <thead><tr><th>Format</th><th>Variant benchmarked</th><th>How Q1-Q7 run</th></tr></thead>
+            <tbody>
+              <tr><td><b>CSV</b> (canonical raw)</td><td>hierarchical file tree + MANIFEST</td><td>python csv module, streaming; Q1 = direct path read</td></tr>
+              <tr><td><b>JSON-LD</b></td><td>FULL (per-coupon files, 588); HYBRID kept for exchange</td><td>ijson event streaming with early exit</td></tr>
+              <tr><td><b>SQLite</b></td><td>single file, composite natural PKs WITHOUT ROWID, enforced FKs</td><td>SQL; precomputed track_summary for Q2-Q4/Q7</td></tr>
+              <tr><td><b>PostgreSQL 16</b></td><td>remote host, range-partitioned bulk tables (12+12), BRIN + b-tree</td><td>SQL; parallel scans (4 workers), track_summary matview</td></tr>
+              <tr><td><b>RDF</b></td><td>pyoxigraph on-disk store, 17,419,713 triples (compact modeling)</td><td>SPARQL; run-in/steady-state derived from raw cycles</td></tr>
+            </tbody>
+          </table>
+        </div>
+      </section>
+
+      <!-- ================= METHODOLOGY ================= -->
+      <section id="method">
+        <h1>Methodology</h1>
+        <p class="sub">The product of this study is its measurements; the protocol is designed so a sloppy number cannot survive it (docs/rules/bench-methodology.md).</p>
+        <div class="grid2">
+          <div class="card">
+            <h3>Measurement protocol</h3>
+            <ul>
+              <li><b>One subprocess per run</b> - RSS / CPU / read bytes attribute to exactly that cell; psutil samples the whole process tree every 50 ms (the venv launcher stub alone would report 4 MB).</li>
+              <li><b>1 warm-up + 3 measured runs</b> per cell; medians reported, min/max kept in raw data.</li>
+              <li><b>Randomized cell order</b> (deterministic shuffle from the corpus seed); order recorded.</li>
+              <li><b>30-minute hard timeout</b> per run; a timeout is recorded as data, never retried.</li>
+              <li><b>Correctness gates speed:</b> every run's rows are compared to the canonical expected results (sorted rows, float tolerance 1e-9); a fast run that fails the checksum is an invalid measurement.</li>
+            </ul>
+          </div>
+          <div class="card">
+            <h3>Fairness and honesty</h3>
+            <ul>
+              <li>Identical result-set contract for all five implementations of each query; no format answers a simplified question.</li>
+              <li>Each format plays its idiomatic strength (CSV direct path read, SQL indexes, matviews) - but no LIMIT/sampling that changes the answer.</li>
+              <li><b>Q5 must scan raw cycles in every format</b>; the PostgreSQL EXPLAIN plan is checked to prove track_summary was not used.</li>
+              <li>Cold-cache pass skipped (Windows host has no page-cache drop) - marked, not faked.</li>
+              <li>PostgreSQL is remote: client wall times include the LAN round-trip; server-side cost captured via pg_stat_statements deltas and Prometheus host windows.</li>
+            </ul>
+          </div>
+        </div>
+        <div class="card">
+          <h3>The seven retrieval scenarios</h3>
+          <table class="data">
+            <thead><tr><th>ID</th><th>Scenario</th><th class="num">Result rows</th></tr></thead>
+            <tbody id="method-queries"></tbody>
+          </table>
+        </div>
+        <div class="card">
+          <h3>Environment (recorded per session)</h3>
+          <pre class="env" id="method-env"></pre>
+        </div>
+      </section>
+
+      <!-- ================= STORAGE ================= -->
+      <section id="storage">
+        <h1>Storage Footprint</h1>
+        <p class="sub">Working footprint is what the query engine reads; archival is the compressed form for shelving. All archival sizes converge to 53-57 MB - the information content is the same, the working overhead is not.</p>
+        <div class="kpis four">
+          <div class="kpi"><div class="l">Most compact</div><div class="v v-good">0.83x</div><div class="s">SQLite: 123 MB, indexes included</div></div>
+          <div class="kpi"><div class="l">PostgreSQL</div><div class="v v-neutral">2.67x</div><div class="s">397 MB live; 34.1% of it is indexes</div></div>
+          <div class="kpi"><div class="l">RDF store</div><div class="v v-bad">11.70x</div><div class="s">1.74 GB for the same data</div></div>
+          <div class="kpi"><div class="l">Best archive</div><div class="v v-good">52.8 MB</div><div class="s">csv tree tar.gz (2.8x compression)</div></div>
+        </div>
+        <div class="card">
+          <h3>Working footprint, measured on the 150 MB corpus</h3>
+          <div id="st-work" class="hbars"></div>
+        </div>
+        <div class="grid2">
+          <div class="card">
+            <h3>Archival (compressed) footprint</h3>
+            <div id="st-arch" class="hbars"></div>
+            <p class="muted" style="margin-top:10px">pg_dump -Fc is internally gzip-compressed; RDF also has a Turtle serialization (596 MB, not the archival form). Hybrid JSON-LD is 1.9 MB of metadata + sourceFile links.</p>
+          </div>
+          <div class="card">
+            <h3>Projected working footprint (linear scaling)</h3>
+            <table class="data">
+              <thead><tr><th>Format</th><th class="num">@600 GB</th><th class="num">@1.2 TB</th><th class="num">@6 TB</th></tr></thead>
+              <tbody id="st-proj"></tbody>
+            </table>
+            <p class="muted" style="margin-top:10px">All projected. Storage scales linearly in data volume; coefficients are the measured ratios above.</p>
+          </div>
+        </div>
+      </section>
+
+      <!-- ================= MEASURED ================= -->
+      <section id="measured">
+        <h1>Measured Results <span class="pill base" style="vertical-align:3px">150 MB corpus &middot; warm cache</span></h1>
+        <p class="sub">Median of 3 isolated runs per cell. Pick a query and a metric; bars are linear with the actual value on each bar. Client peak RSS below ~50 ms of runtime under-reports (sampling cadence is spec-fixed at 50 ms) - wall times are exact.</p>
+        <div class="tabs" id="mq-tabs"></div>
+        <div class="tabs" id="mm-tabs" style="margin-left:10px"></div>
+        <div class="card">
+          <h3 id="m-title"></h3>
+          <div id="m-bars" class="hbars"></div>
+        </div>
+        <div class="card">
+          <h3>Full median matrix - wall time</h3>
+          <table class="data"><thead id="m-mat-h"></thead><tbody id="m-mat"></tbody></table>
+        </div>
+        <div class="card">
+          <h3>PostgreSQL server side (remote host, 4 cores / 3.8 GiB)</h3>
+          <table class="data">
+            <thead><tr><th>Query</th><th class="num">Server exec per call</th><th class="num">Shared blocks hit</th><th class="num">Read from disk</th></tr></thead>
+            <tbody id="m-pgsrv"></tbody>
+          </table>
+          <p class="muted" style="margin-top:10px">pg_stat_statements deltas over the session (4 calls per query). Zero disk reads: the whole 397 MB database sits in the 1 GB shared_buffers. Host peak CPU during the session: 9.2%; RAM delta ~5 MB - the server is nowhere near its limits at this scale.</p>
+        </div>
+      </section>
+
+      <!-- ================= PROJECTIONS ================= -->
+      <section id="proj">
+        <h1>Projections <span class="pill imp" style="vertical-align:3px">every number here is projected, not measured</span></h1>
+        <p class="sub">Per-cell scaling laws calibrated on the measured 150 MB point after subtracting each format's harness floor (interpreter start, imports, connection). Flags: IMPRACTICAL over 1 hour, FAIL over 24 hours. Pick a target scale.</p>
+        <div class="tabs" id="p-tabs"></div>
+        <div class="card">
+          <h3 id="p-title"></h3>
+          <table class="data"><thead id="p-mat-h"></thead><tbody id="p-mat"></tbody></table>
+          <p class="muted" style="margin-top:10px" id="p-note"></p>
+        </div>
+        <div class="grid2">
+          <div class="card">
+            <h3>Scaling law per access pattern</h3>
+            <table class="data">
+              <thead><tr><th>Law</th><th>Growth</th><th>Applied to</th></tr></thead>
+              <tbody>
+                <tr><td class="mono">O(1)</td><td>flat</td><td>CSV / JSON-LD Q1 (one file, size fixed)</td></tr>
+                <tr><td class="mono">O(log n)</td><td>index depth</td><td>SQLite / PostgreSQL / RDF Q1</td></tr>
+                <tr><td class="mono">O(log n + k)</td><td>~linear (result set grows)</td><td>relational + RDF Q2, Q4</td></tr>
+                <tr><td class="mono">O(k log n)</td><td>linear x depth</td><td>relational Q3, Q6; RDF Q6</td></tr>
+                <tr><td class="mono">O(n)</td><td>linear, 1 core</td><td>all CSV/JSON scans; SQLite Q5/Q7; RDF Q3/Q5/Q7</td></tr>
+                <tr><td class="mono">O(n / cores)</td><td>linear / parallel workers</td><td>PostgreSQL Q5, Q7 (partitioned scan)</td></tr>
+              </tbody>
+            </table>
+          </div>
+          <div class="card">
+            <h3>Full-scan degradation (Q5 + Q7 geometric mean)</h3>
+            <div id="p-degr"></div>
+            <p class="muted" style="margin-top:8px">First column measured, the rest projected. PostgreSQL rides its parallel partitioned scan; single-threaded engines grow linearly.</p>
+          </div>
+        </div>
+        <div class="callout">PostgreSQL is the only format with all seven queries below one hour at every scale - Q5 at 6 TB projects to 37 minutes on 16 cores. SQLite holds to 600 GB, then its single-threaded scans pass the hour mark. JSON-LD full scans reach 4.6 days at 6 TB; RDF Q5 reaches 27.8 days.</div>
+      </section>
+
+      <!-- ================= HARDWARE ================= -->
+      <section id="hw">
+        <h1>Hardware &amp; Cost <span class="pill base" style="vertical-align:3px">street prices as of 2026-07-11</span></h1>
+        <p class="sub">Bill-of-materials costing: whole 64 GB RDIMM modules, whole 7.68 TB NVMe drives, a priced 1U chassis; disk includes 30% free-space headroom. RAM model: relational hot set = 10% of index size + 4 GB working set; RDF hot set = 20% of the store; streaming formats are flat. Single-node RAM ceiling: 1 TB.</p>
+        <div class="tabs" id="h-tabs"></div>
+        <div class="card">
+          <h3 id="h-title"></h3>
+          <table class="data">
+            <thead><tr><th>Format</th><th>Exact configuration</th><th class="num">Chassis</th><th class="num">RAM</th><th class="num">Disk</th><th class="num">Extra nodes</th><th class="num">Total</th></tr></thead>
+            <tbody id="h-bom"></tbody>
+          </table>
+        </div>
+        <div class="grid2">
+          <div class="card">
+            <h3>Total cost comparison</h3>
+            <div id="h-bars" class="hbars"></div>
+          </div>
+          <div class="card">
+            <h3>Component prices &amp; sources <span class="pill base">retrieved 2026-07-11</span></h3>
+            <ul class="src">
+              <li>64 GB DDR5-5600 ECC RDIMM (A-Tech, 2x32GB): <b>$1,887</b><br><a href="https://atechmemory.com/collections/ddr5-memory-ram">atechmemory.com/collections/ddr5-memory-ram</a></li>
+              <li>7.68 TB NVMe U.2 1DWPD enterprise SSD (Cloud Ninjas): <b>$3,995</b><br><a href="https://cloudninjas.com/products/new-7-68tb-nvme-u-2-1dwpd-sie-2-5-enterprise-solid-state-drive-for-14th-15th-16th-gen-dell">cloudninjas.com/products/new-7-68tb-nvme-u-2-1dwpd...</a></li>
+              <li>1U AMD EPYC chassis (Broadberry starting configs): AS-1015CS-TNR <b>$4,618</b>, AS-1115CS-TNR <b>$6,272</b><br><a href="https://www.broadberry.com/amd-epyc-9004-supermicro-servers">broadberry.com/amd-epyc-9004-supermicro-servers</a></li>
+              <li>Market context - server DRAM 2025-2026 price surge:<br><a href="https://www.techpowerup.com/342331/server-dram-pricing-jumps-50-only-70-of-orders-getting-filled">techpowerup.com/342331/server-dram-pricing-jumps-50...</a></li>
+            </ul>
+            <p class="muted" style="margin-top:10px">Spot prices during a documented DRAM surge; re-cost by editing out/config/hw_prices.yaml and re-running tasks 10-11.</p>
+          </div>
+        </div>
+        <div class="card">
+          <h3>Cost vs speed at 600 GB (projected)</h3>
+          <div id="h-scatter"></div>
+        </div>
+      </section>
+
+      <!-- ================= SCORING ================= -->
+      <section id="score">
+        <h1>Scoring &amp; Verdict</h1>
+        <p class="sub">Subscores are 10 x best/value per metric, weighted: search speed x5 (0-50, inverse geometric mean of Q1-Q7), RAM economy x2 (0-20), disk economy x1 (0-10); maximum 80. A FAIL projection at 600 GB zeroes the search subscore.</p>
+        <div class="grid2">
+          <div class="card">
+            <h3>Measured scale (150 MB)</h3>
+            <div class="legend"><span><i style="background:#1f5fdb"></i>search x5</span><span><i style="background:#e0892e"></i>RAM x2</span><span><i style="background:#0f9d8a"></i>disk x1</span></div>
+            <div id="sc-measured"></div>
+            <p class="muted" style="margin-top:8px">At laptop scale SQLite is the honest winner: near-instant indexed queries in one file with the smallest footprint.</p>
+          </div>
+          <div class="card">
+            <h3>Projected at 600 GB</h3>
+            <div class="legend"><span><i style="background:#1f5fdb"></i>search x5</span><span><i style="background:#e0892e"></i>RAM x2</span><span><i style="background:#0f9d8a"></i>disk x1</span></div>
+            <div id="sc-proj"></div>
+            <p class="muted" style="margin-top:8px">At production scale the ranking flips: parallel indexed retrieval dominates the weights, and PostgreSQL takes the lead.</p>
+          </div>
+        </div>
+        <div class="card">
+          <h3>Use-case mapping</h3>
+          <table class="data">
+            <thead><tr><th>Use case</th><th>Recommended format</th></tr></thead>
+            <tbody>
+              <tr><td>Interactive analysis (joins, aggregations)</td><td><b>PostgreSQL</b>; SQLite acceptable single-user up to ~600 GB</td></tr>
+              <tr><td>Report generation (repeated summaries)</td><td><b>PostgreSQL</b> (track_summary materialized view)</td></tr>
+              <tr><td>Search / filtering</td><td><b>PostgreSQL or SQLite</b> (indexed); flat formats need full scans</td></tr>
+              <tr><td>Archiving</td><td><b>CSV tree + tar.gz</b> (canonical raw) plus pg_dump of the system of record</td></tr>
+              <tr><td>Inter-lab exchange</td><td><b>Hybrid JSON-LD</b> (metadata + sourceFile links to CSV)</td></tr>
+              <tr><td>Semantic / ontology queries</td><td><b>Virtual RDF layer</b> over PostgreSQL (e.g. Ontop OBDA), not a materialized triplestore</td></tr>
+            </tbody>
+          </table>
+        </div>
+        <div class="card">
+          <h3>Limitations</h3>
+          <ul>
+            <li>Single-node measurements on one Windows host + one remote PostgreSQL host; no cluster variance.</li>
+            <li>Simulated corpus (physically plausible, fixed seed); real instrument data may distribute differently.</li>
+            <li>Extrapolation is analytic, calibrated on one 150 MB point; no intermediate-scale validation runs.</li>
+            <li>Warm-cache only (no page-cache drop on Windows); harness floor assumed constant across scales.</li>
+            <li>RDF numbers reflect compact modeling and client-side derivation of run-in/steady-state; a SPARQL-side aggregation engine could shift, not remove, the scan penalty.</li>
+            <li>Hardware prices are spot street prices (2026-07-11) during a DRAM price surge.</li>
+          </ul>
+        </div>
+        <div class="foot">
+          Generated from the LabDataStorageEvaluation pipeline artifacts (run of 2026-07-11, seed 20260711): out/report/REPORT.md, out/bench/results_median.csv, extrapolation.csv, hardware_sizing.csv, storage_sizes.csv, out/config/hw_prices.yaml. The full decision document with charts is out/report/REPORT.md; this page is the interactive companion. Related study: <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_FK_Architecture.html" target="_blank" rel="noopener">Foreign-Key Architecture Study</a>.
+        </div>
+      </section>
+
+      <!-- ================= ABOUT ================= -->
+      <section id="about">
+        <h1>About</h1>
+        <p class="sub">Attribution and provenance for this document. It is distributed as a single standalone HTML file; everything needed to verify or reproduce the numbers is referenced below.</p>
+        <div class="card">
+          <h3>Author</h3>
+          <p>This study was conducted by <b>Mary Goncharenko</b>, PhD student at the Tribology Laboratory, University of Florida, Gainesville, FL, during a research internship at the tribology laboratory of Sandia National Laboratories, with an assignment on laboratory ontology development.</p>
+          <p>The research was carried out with the technical assistance of her father, <b>Vasiliy Goncharenko</b>, Chief Technology Architect at SoftCreator, LLC (enterprise data architecture across relational and semantic stores; 35 years in technology).</p>
+          <p class="muted">Profiles: <a class="ext" href="https://www.linkedin.com/in/marygoncharenko/" target="_blank" rel="noopener">linkedin.com/in/marygoncharenko</a> &middot; <a class="ext" href="https://linkedin.com/in/vasiliy-goncharenko" target="_blank" rel="noopener">linkedin.com/in/vasiliy-goncharenko</a></p>
+        </div>
+        <div class="card">
+          <h3>Source repository</h3>
+          <p>The complete evaluation pipeline - task specifications, binding conventions, source code, and the generated report - is publicly available at:</p>
+          <p class="mono"><a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">https://git.boskadoff.com/Public/LabDataStorageEvaluation</a></p>
+          <p>This page is the interactive companion to the repository's <span class="mono">out/report/REPORT.md</span>; all numbers embedded here come from the pipeline run of 2026-07-11. The corpus regenerates byte-identically from seed 20260711 (MANIFEST.csv sha256 proof), so every figure is independently reproducible from the repository alone.</p>
+        </div>
+        <div class="card">
+          <h3>Data statement</h3>
+          <p>The measurement corpus is simulated: physically plausible models of a Pt-Au LDRD tribology study context at Sandia National Laboratories, generated from a fixed random seed. It contains no experimental measurements and no export-controlled data. Hardware prices cited in the cost model are public street prices retrieved 2026-07-11 (sources in the Hardware &amp; Cost section).</p>
+          <p>Companion pages in the same repository: <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_Storage_Decision.html" target="_blank" rel="noopener">Storage Decision (one-page executive summary)</a> &middot; <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation/raw/branch/master/docs/research/Tribology_FK_Architecture.html" target="_blank" rel="noopener">Foreign-Key Architecture Study</a>.</p>
+        </div>
+        <div class="foot">Standalone distributable file. If you received this document outside the repository, the canonical source is <a class="ext" href="https://git.boskadoff.com/Public/LabDataStorageEvaluation" target="_blank" rel="noopener">git.boskadoff.com/Public/LabDataStorageEvaluation</a>.</div>
+      </section>
+
+    </main>
+  </div>
+</div>
+
+<script>
+"use strict";
+
+/* ---------- embedded data (from the 2026-07-11 pipeline run) ---------- */
+const FMT = ["csv","json","sqlite","pg","rdf"];
+const NAME = {csv:"CSV", json:"JSON-LD", sqlite:"SQLite", pg:"PostgreSQL", rdf:"RDF triplestore"};
+const COLOR = {csv:"#9aa6b6", json:"#e0892e", sqlite:"#0f9d8a", pg:"#1f5fdb", rdf:"#c0476b"};
+const QN = [1,2,3,4,5,6,7];
+const QDESC = {
+  1:["COF vs cycle curve for one track (B722-W2-C13-T2)",1000],
+  2:["Steady-state COF for coupons with mean Au = 10 +/- 0.5 wt%",33],
+  3:["Hardness vs steady-state COF across all batches",480],
+  4:["Anomaly filter: dry N2, 100 mN, cof_ss > 0.20",469],
+  5:["Mean run-in cycles per batch over ALL raw cycles (no summaries)",4],
+  6:["Wear volume vs load per coupon",480],
+  7:["Stribeck-style mean COF by (environment, speed x load)",2]
+};
+
+/* measured medians, per format: [q1..q7] */
+const WALL = {
+  csv:[0.06,0.1661,1.1521,0.715,1.0948,0.2756,1.2735],
+  json:[0.1113,1.4788,9.9408,9.9218,9.9623,9.7927,10.0781],
+  sqlite:[0.0617,0.0607,0.1134,0.0604,0.7659,0.0612,0.7094],
+  pg:[0.2214,0.2206,0.2211,0.2232,0.4405,0.2212,0.3342],
+  rdf:[0.1131,2.8054,46.1273,19.8245,59.5884,0.1682,40.5534]
+};
+const RSS = {
+  csv:[3.9,20.5,21.1,20.9,20.7,20.2,21.0],
+  json:[22.2,23.9,24.4,24.5,24.4,24.1,24.1],
+  sqlite:[3.9,3.9,4.1,3.9,25.0,3.9,25.0],
+  pg:[42.2,42.1,42.0,41.9,42.0,41.6,42.1],
+  rdf:[31.6,85.4,263.9,167.9,264.0,45.9,263.6]
+};
+const READ = {
+  csv:[0,2.45,31.76,16.82,31.78,1.23,32.47],
+  json:[1.4,53.08,297.57,297.75,297.57,297.82,297.94],
+  sqlite:[0,0,0,0,117.91,0,92.97],
+  pg:[4.08,4.08,4.08,4.08,4.08,4.08,4.08],
+  rdf:[3.62,375.56,5445.1,2694.56,5390.46,15.92,5380.51]
+};
+
+/* projections: scale -> format -> [[seconds,flag] x7]; flag "" | "I" | "F" */
+const PROJ = {
+  600: {
+    csv:[[0.06,""],[428.301,""],[4408.0,"I"],[2643.775,""],[4176.725,"I"],[870.266,""],[4897.995,"I"]],
+    json:[[0.111,""],[5519.622,"I"],[39673.993,"I"],[39597.306,"I"],[39760.772,"I"],[39076.231,"I"],[40228.164,"I"]],
+    sqlite:[[0.062,""],[1.271,""],[339.238,""],[0.06,""],[2847.603,""],[5.18,""],[2619.558,""]],
+    pg:[[0.222,""],[0.221,""],[3.42,""],[10.715,""],[222.111,""],[4.06,""],[114.849,""]],
+    rdf:[[0.113,""],[10866.789,"I"],[185722.881,"F"],[79559.375,"I"],[240054.647,"F"],[352.729,""],[163225.475,"F"]]
+  },
+  1200: {
+    csv:[[0.06,""],[856.543,""],[8815.94,"I"],[5287.489,"I"],[8353.391,"I"],[1740.472,""],[9795.931,"I"]],
+    json:[[0.111,""],[11039.133,"I"],[79347.876,"I"],[79194.5,"I"],[79521.432,"I"],[78152.351,"I"],[80456.218,"I"]],
+    sqlite:[[0.063,""],[2.482,""],[699.328,""],[0.06,""],[5695.146,"I"],[10.615,""],[5239.055,"I"]],
+    pg:[[0.222,""],[0.221,""],[6.817,""],[21.209,""],[444.001,""],[8.137,""],[229.477,""]],
+    rdf:[[0.113,""],[21733.465,"I"],[371445.649,"F"],[159118.636,"F"],[480109.181,"F"],[727.088,""],[326450.837,"F"]]
+  },
+  6000: {
+    csv:[[0.06,""],[4282.474,"I"],[44079.46,"I"],[26437.206,"I"],[41766.715,"I"],[8702.119,"I"],[48979.414,"I"]],
+    json:[[0.111,""],[55195.222,"I"],[396738.933,"F"],[395972.054,"F"],[397606.717,"F"],[390761.312,"F"],[402280.643,"F"]],
+    sqlite:[[0.063,""],[12.169,""],[3739.195,"I"],[0.06,""],[28475.49,"I"],[56.5,""],[26195.034,"I"]],
+    pg:[[0.222,""],[0.221,""],[35.495,""],[105.162,""],[2219.125,""],[42.55,""],[1146.503,""]],
+    rdf:[[0.113,""],[108666.874,"F"],[1857227.792,"F"],[795592.727,"F"],[2400545.453,"F"],[3887.402,"I"],[1632253.733,"F"]]
+  }
+};
+
+/* storage bytes */
+const WORK = {csv:148654472, json:311300222, sqlite:123400192, pg:397467648, rdf:1739414085};
+const ARCH = {csv:[52827471,"tar.gz of the tree"], json:[57162674,"tar.gz (full + hybrid)"], sqlite:[55149977,"gzip of tribo.db"], pg:[53054380,"pg_dump -Fc"], rdf:[126487650,"N-Triples gz"]};
+const COEFF = {csv:1.0, json:2.09, sqlite:0.83, pg:2.67, rdf:11.70};
+
+/* hardware sizing: scale -> format -> {modules,drives,nodes,cores,base,ram,disk,extra,total} */
+const HW = {
+  600: {
+    csv:{m:1,d:1,n:1,c:16,base:4618,ram:1887,disk:3995,extra:0,total:10500},
+    json:{m:1,d:1,n:1,c:16,base:4618,ram:1887,disk:3995,extra:0,total:10500},
+    sqlite:{m:1,d:1,n:1,c:16,base:4618,ram:1887,disk:3995,extra:0,total:10500},
+    pg:{m:1,d:1,n:1,c:16,base:4618,ram:1887,disk:3995,extra:0,total:10500},
+    rdf:{m:22,d:2,n:2,c:16,base:4618,ram:41514,disk:7990,extra:4618,total:58740}
+  },
+  1200: {
+    csv:{m:1,d:1,n:1,c:16,base:4618,ram:1887,disk:3995,extra:0,total:10500},
+    json:{m:1,d:1,n:1,c:16,base:4618,ram:1887,disk:3995,extra:0,total:10500},
+    sqlite:{m:1,d:1,n:1,c:16,base:4618,ram:1887,disk:3995,extra:0,total:10500},
+    pg:{m:2,d:1,n:1,c:16,base:4618,ram:3774,disk:3995,extra:0,total:12387},
+    rdf:{m:44,d:3,n:3,c:16,base:4618,ram:83028,disk:11985,extra:9236,total:108867}
+  },
+  6000: {
+    csv:{m:1,d:2,n:1,c:16,base:4618,ram:1887,disk:7990,extra:0,total:14495},
+    json:{m:1,d:3,n:1,c:16,base:4618,ram:1887,disk:11985,extra:0,total:18490},
+    sqlite:{m:3,d:1,n:1,c:16,base:4618,ram:5661,disk:3995,extra:0,total:14274},
+    pg:{m:9,d:3,n:1,c:16,base:4618,ram:16983,disk:11985,extra:0,total:33586},
+    rdf:{m:220,d:12,n:14,c:16,base:4618,ram:415140,disk:47940,extra:60034,total:527732}
+  }
+};
+
+/* weighted scores: [search, ram, disk, total] */
+const SCORES = {
+  measured:{csv:[15.5,20.0,8.3,43.8], json:[1.7,17.2,4.0,22.9], sqlite:[50.0,16.9,10.0,76.9], pg:[26.2,10.0,3.1,39.3], rdf:[1.3,1.6,0.7,3.6]},
+  proj:{csv:[0.6,20.0,8.3,28.9], json:[0.1,20.0,4.0,24.0], sqlite:[21.8,15.2,10.0,47.0], pg:[50.0,5.5,3.1,58.6], rdf:[0.0,0.2,0.7,0.9]}
+};
+
+/* pg server per-call exec ms + shared blocks (session deltas / 4 calls) */
+const PGSRV = {1:[0.42,72],2:[0.43,1312],3:[4.92,6500],4:[1.09,140],5:[237.26,189692],6:[1.68,172],7:[107.14,32456]};
+
+const ENVIRONMENT = [
+"benchmark_host_os: Windows-2025Server-10.0.26100-SP0",
+"benchmark_host_cpu: Intel64 Family 6 Model 141 (4 cores physical/logical), 8 GiB RAM",
+"benchmark_host_disks: VMware Virtual NVMe Disk / SSD / 100 GB",
+"python: 3.12.10 &middot; sqlite 3.49.1 &middot; psycopg 3.3.4 &middot; pyoxigraph 0.5.9 &middot; ijson 3.5.1 &middot; psutil 7.2.2",
+"postgresql_server: PostgreSQL 16.14 (Ubuntu 24.04), remote host, 4 cores / 3.8 GiB",
+"pg tuning: shared_buffers=1GB &middot; work_mem=32MB &middot; effective_cache_size=2500MB &middot; max_parallel_workers_per_gather=4",
+"cold_cache_pass: skipped (Windows host, no page-cache drop)",
+"pg_backend_sampling: remote server - Prometheus windows + pg_stat_statements deltas instead"
+].join("\n");
+
+/* ---------- formatting helpers ---------- */
+function fmtTime(s){
+  if(s < 10) return s.toFixed(2) + " s";
+  if(s < 120) return s.toFixed(0) + " s";
+  if(s < 7200) return (s/60).toFixed(1) + " min";
+  if(s < 172800) return (s/3600).toFixed(1) + " h";
+  return (s/86400).toFixed(1) + " d";
+}
+function fmtSize(b){
+  if(b >= 1e12) return (b/1e12).toFixed(2) + " TB";
+  if(b >= 1e9) return (b/1e9).toFixed(2) + " GB";
+  return (b/1e6).toFixed(0) + " MB";
+}
+function fmtMoney(v){ return "$" + v.toLocaleString("en-US"); }
+function esc(t){ return String(t).replace(/&/g,"&amp;").replace(/</g,"&lt;"); }
+function el(id){ return document.getElementById(id); }
+function flagPill(f){
+  if(f === "F") return ' <span class="pill fail">FAIL</span>';
+  if(f === "I") return ' <span class="pill imp">&gt; 1 h</span>';
+  return "";
+}
+function geomean(a){ return Math.exp(a.reduce((s,v)=>s+Math.log(v),0)/a.length); }
+
+/* one horizontal bar row; linear scale against max */
+function barRow(label, sub, frac, color, val){
+  return '<div class="hbar"><div class="lab">' + label + (sub ? '<small>' + sub + '</small>' : '') + '</div>' +
+    '<div class="trk"><div class="fill" style="width:' + Math.max(0.4, frac*100).toFixed(2) + '%;background:' + color + '"></div></div>' +
+    '<div class="val">' + val + '</div></div>';
+}
+function tabs(containerId, items, onPick){
+  const box = el(containerId);
+  box.innerHTML = items.map((t,i) => '<button data-i="' + i + '"' + (i===0 ? ' class="on"' : '') + '>' + t + '</button>').join("");
+  box.querySelectorAll("button").forEach(b => b.addEventListener("click", () => {
+    box.querySelectorAll("button").forEach(x => x.classList.remove("on"));
+    b.classList.add("on");
+    onPick(parseInt(b.dataset.i,10));
+  }));
+}
+
+/* ---------- navigation ---------- */
+/* menu items only: the sidebar footer also holds a regular external link */
+const navLinks = document.querySelectorAll('nav.side a[href^="#"]');
+function show(id){
+  if(!document.getElementById(id)) return;
+  document.querySelectorAll("main section").forEach(s => s.classList.toggle("on", s.id === id));
+  navLinks.forEach(a => a.classList.toggle("active", a.getAttribute("href") === "#" + id));
+  el("main").scrollTop = 0;
+}
+navLinks.forEach(a => a.addEventListener("click", e => {
+  e.preventDefault();
+  const id = a.getAttribute("href").slice(1);
+  show(id);
+  history.replaceState(null, "", "#" + id);
+}));
+if(location.hash && document.getElementById(location.hash.slice(1))) show(location.hash.slice(1));
+
+/* ---------- dashboard ---------- */
+(function(){
+  const rows = FMT.map(f => {
+    const g150 = geomean(WALL[f]);
+    const g600 = geomean(PROJ[600][f].map(x => x[0]));
+    return [NAME[f], "geometric mean of Q1-Q7", g150, fmtTime(g150) + '  <span class="muted">(' + fmtTime(g600) + ' @600 GB)</span>', COLOR[f]];
+  });
+  const max = Math.max(...rows.map(r => r[2]));
+  el("dash-geo").innerHTML = rows.map(r => barRow(r[0], r[1], r[2]/max, r[4], r[3])).join("");
+
+  const breaks = {
+    csv:["Direct path reads; canonical archive; best compression","Full scans pass 1 h just above 600 GB"],
+    json:["Self-describing exchange files; hybrid variant is 1.9 MB","Every scan reads the whole corpus: hours at 600 GB"],
+    sqlite:["Fastest at 150 MB; one file; smallest footprint","Single-threaded scans pass 1 h at ~1.2 TB"],
+    pg:["Parallel partitioned scans + matviews; flat 0.2-0.4 s today","None found up to 6 TB (Q5 = 37 min)"],
+    rdf:["Native semantics, SPARQL","11.7x disk, 20% hot set in RAM, scans measured in days"]
+  };
+  el("dash-breaks").innerHTML = FMT.map(f => {
+    const fl = PROJ[600][f];
+    const nI = fl.filter(x => x[1]==="I").length, nF = fl.filter(x => x[1]==="F").length;
+    let flags = (nI+nF)===0 ? '<span class="pill ok">all 7 OK</span>' :
+      (nI ? '<span class="pill imp">' + nI + ' x &gt;1 h</span> ' : '') + (nF ? '<span class="pill fail">' + nF + ' x FAIL</span>' : '');
+    return "<tr><td><b>" + NAME[f] + "</b></td><td>" + breaks[f][0] + "</td><td>" + breaks[f][1] + "</td><td>" + flags + "</td></tr>";
+  }).join("");
+})();
+
+/* ---------- methodology ---------- */
+el("method-queries").innerHTML = QN.map(q =>
+  '<tr><td class="mono">Q' + q + '</td><td>' + QDESC[q][0] + '</td><td class="num">' + QDESC[q][1].toLocaleString("en-US") + '</td></tr>').join("");
+el("method-env").innerHTML = ENVIRONMENT;
+
+/* ---------- storage ---------- */
+(function(){
+  const workMax = WORK.rdf;
+  el("st-work").innerHTML = FMT.map(f =>
+    barRow(NAME[f], COEFF[f].toFixed(2) + "x vs raw CSV", WORK[f]/workMax, COLOR[f], fmtSize(WORK[f]))).join("");
+  const archMax = ARCH.rdf[0];
+  el("st-arch").innerHTML = FMT.map(f =>
+    barRow(NAME[f], ARCH[f][1], ARCH[f][0]/archMax, COLOR[f], fmtSize(ARCH[f][0]))).join("");
+  el("st-proj").innerHTML = FMT.map(f =>
+    "<tr><td><b>" + NAME[f] + "</b></td>" + [600,1200,6000].map(s =>
+      '<td class="num">' + fmtSize(COEFF[f]*s*1e9) + "</td>").join("") + "</tr>").join("");
+})();
+
+/* ---------- measured results ---------- */
+(function(){
+  const METRICS = [
+    ["Wall time", WALL, fmtTime],
+    ["Client peak RSS", RSS, v => v.toFixed(0) + " MB"],
+    ["Read bytes", READ, v => v >= 1000 ? (v/1000).toFixed(2) + " GB" : v.toFixed(1) + " MB"]
+  ];
+  let q = 0, m = 0;
+  function render(){
+    const [mname, data, fmt] = METRICS[m];
+    el("m-title").textContent = "Q" + QN[q] + " - " + QDESC[QN[q]][0] + " - " + mname.toLowerCase() + " (median of 3 runs)";
+    const max = Math.max(...FMT.map(f => data[f][q])) || 1;
+    el("m-bars").innerHTML = FMT.map(f =>
+      barRow(NAME[f], "", (data[f][q])/max, COLOR[f], fmt(data[f][q]))).join("");
+  }
+  tabs("mq-tabs", QN.map(n => "Q" + n), i => { q = i; render(); });
+  tabs("mm-tabs", METRICS.map(x => x[0]), i => { m = i; render(); });
+  render();
+
+  el("m-mat-h").innerHTML = "<tr><th>Format</th>" + QN.map(n => '<th class="num">Q' + n + "</th>").join("") + "</tr>";
+  el("m-mat").innerHTML = FMT.map(f =>
+    "<tr><td><b>" + NAME[f] + "</b></td>" + WALL[f].map(v => '<td class="num">' + fmtTime(v) + "</td>").join("") + "</tr>").join("");
+
+  el("m-pgsrv").innerHTML = QN.map(n =>
+    '<tr><td class="mono">Q' + n + "</td><td class=\"num\">" + PGSRV[n][0].toFixed(2) + " ms</td><td class=\"num\">" +
+    PGSRV[n][1].toLocaleString("en-US") + '</td><td class="num">0</td></tr>').join("");
+})();
+
+/* ---------- projections ---------- */
+(function(){
+  const SCALES = [600, 1200, 6000];
+  const SLBL = {600:"600 GB", 1200:"1.2 TB", 6000:"6 TB"};
+  function render(scale){
+    el("p-title").textContent = "Projected wall times at " + SLBL[scale] + " of raw data";
+    el("p-mat-h").innerHTML = "<tr><th>Format</th>" + QN.map(n => '<th class="num">Q' + n + "</th>").join("") + "</tr>";
+    el("p-mat").innerHTML = FMT.map(f =>
+      "<tr><td><b>" + NAME[f] + "</b></td>" + PROJ[scale][f].map(x =>
+        '<td class="num">' + fmtTime(x[0]) + flagPill(x[1]) + "</td>").join("") + "</tr>").join("");
+    const total = FMT.reduce((acc,f) => {
+      PROJ[scale][f].forEach(x => { if(x[1]==="I") acc.i++; if(x[1]==="F") acc.f++; });
+      return acc;
+    }, {i:0, f:0});
+    el("p-note").textContent = "Flags at this scale: " + total.i + " x IMPRACTICAL (over 1 hour), " + total.f +
+      " x FAIL (over 24 hours), " + (35 - total.i - total.f) + " x OK. Scaling ratio vs the measured corpus: x" +
+      Math.round(scale*1e9/WORK.csv).toLocaleString("en-US") + ".";
+  }
+  tabs("p-tabs", SCALES.map(s => SLBL[s]), i => render(SCALES[i]));
+  render(600);
+
+  /* degradation mini-table: full scan class geometric mean */
+  const cols = [["measured 150 MB", f => geomean([WALL[f][4], WALL[f][6]])]]
+    .concat(SCALES.map(s => ["projected " + SLBL[s], f => geomean([PROJ[s][f][4][0], PROJ[s][f][6][0]])]));
+  let html = '<table class="data"><thead><tr><th>Format</th>' +
+    cols.map(c => '<th class="num">' + c[0] + "</th>").join("") + "</tr></thead><tbody>";
+  html += FMT.map(f => "<tr><td><b>" + NAME[f] + "</b></td>" +
+    cols.map(c => '<td class="num">' + fmtTime(c[1](f)) + "</td>").join("") + "</tr>").join("");
+  html += "</tbody></table>";
+  el("p-degr").innerHTML = html;
+})();
+
+/* ---------- hardware ---------- */
+(function(){
+  const SCALES = [600, 1200, 6000];
+  const SLBL = {600:"600 GB", 1200:"1.2 TB", 6000:"6 TB"};
+  function config(h){
+    return h.n + " x 1U Supermicro AS-1015CS-TNR (" + h.c + " cores) &middot; " +
+      h.m + " x 64 GB DDR5 ECC RDIMM &middot; " + h.d + " x 7.68 TB NVMe U.2";
+  }
+  function render(scale){
+    el("h-title").textContent = "Exact configuration and cost breakdown at " + SLBL[scale] + " (projected sizing)";
+    el("h-bom").innerHTML = FMT.map(f => {
+      const h = HW[scale][f];
+      return "<tr><td><b>" + NAME[f] + "</b></td><td>" + config(h) + "</td>" +
+        [h.base, h.ram, h.disk, h.extra, h.total].map((v,i) =>
+          '<td class="num"' + (i===4 ? ' style="font-weight:600"' : '') + ">" + fmtMoney(v) + "</td>").join("") + "</tr>";
+    }).join("");
+    const max = Math.max(...FMT.map(f => HW[scale][f].total));
+    el("h-bars").innerHTML = FMT.map(f => {
+      const h = HW[scale][f];
+      return barRow(NAME[f], h.n > 1 ? h.n + " nodes" : "single node", h.total/max, COLOR[f], fmtMoney(h.total));
+    }).join("");
+    scatter(scale);
+  }
+  function scatter(scale){
+    const W = 860, H = 340, L = 70, R = 30, T = 20, B = 46;
+    const pts = FMT.map(f => ({f, cost: HW[scale][f].total, geo: geomean(PROJ[scale][f].map(x => x[0]))}));
+    const maxC = Math.max(...pts.map(p => p.cost)) * 1.15;
+    const maxG = Math.max(...pts.map(p => p.geo)) * 1.15;
+    const X = c => L + (W-L-R) * c / maxC;
+    const Y = g => T + (H-T-B) * (1 - g / maxG);
+    let s = '<svg viewBox="0 0 ' + W + ' ' + H + '" style="width:100%;max-width:900px">';
+    /* axes + linear ticks in real units */
+    for(let i = 1; i <= 4; i++){
+      const c = maxC * i / 4, x = X(c);
+      s += '<line x1="' + x + '" y1="' + T + '" x2="' + x + '" y2="' + (H-B) + '" stroke="#eef0f2"/>';
+      s += '<text x="' + x + '" y="' + (H-B+18) + '" font-size="11" fill="#838a95" text-anchor="middle">$' + Math.round(c/1000) + 'k</text>';
+      const g = maxG * i / 4, y = Y(g);
+      s += '<line x1="' + L + '" y1="' + y + '" x2="' + (W-R) + '" y2="' + y + '" stroke="#eef0f2"/>';
+      s += '<text x="' + (L-8) + '" y="' + (y+4) + '" font-size="11" fill="#838a95" text-anchor="end">' + fmtTime(g) + '</text>';
+    }
+    s += '<line x1="' + L + '" y1="' + (H-B) + '" x2="' + (W-R) + '" y2="' + (H-B) + '" stroke="#cfd4da"/>';
+    s += '<line x1="' + L + '" y1="' + T + '" x2="' + L + '" y2="' + (H-B) + '" stroke="#cfd4da"/>';
+    s += '<text x="' + ((L+W-R)/2) + '" y="' + (H-6) + '" font-size="11.5" fill="#5b6472" text-anchor="middle">estimated hardware cost (bill of materials, linear)</text>';
+    s += '<text x="14" y="' + ((T+H-B)/2) + '" font-size="11.5" fill="#5b6472" text-anchor="middle" transform="rotate(-90 14 ' + ((T+H-B)/2) + ')">typical query time, projected (linear)</text>';
+    /* better = lower-left cue */
+    s += '<text x="' + (L+16) + '" y="' + (H-B-10) + '" font-size="11.5" fill="#0f7a45" font-weight="600">better: lower-left</text>';
+    const OFF = {csv:[12,-8], json:[12,4], sqlite:[12,14], pg:[12,-6], rdf:[-8,-12]};
+    const ANCH = {csv:"start", json:"start", sqlite:"start", pg:"start", rdf:"end"};
+    pts.forEach(p => {
+      s += '<circle cx="' + X(p.cost) + '" cy="' + Y(p.geo) + '" r="7" fill="' + COLOR[p.f] + '"/>';
+      s += '<text x="' + (X(p.cost)+OFF[p.f][0]) + '" y="' + (Y(p.geo)+OFF[p.f][1]) + '" font-size="11.5" fill="#3d424b" text-anchor="' + ANCH[p.f] + '">' +
+        NAME[p.f] + ': ' + fmtMoney(p.cost) + ', ' + fmtTime(p.geo) + '</text>';
+    });
+    s += "</svg>";
+    el("h-scatter").innerHTML = s;
+  }
+  tabs("h-tabs", SCALES.map(s => SLBL[s]), i => render(SCALES[i]));
+  render(600);
+})();
+
+/* ---------- scoring ---------- */
+function renderScores(containerId, board){
+  const order = FMT.slice().sort((a,b) => board[b][3] - board[a][3]);
+  el(containerId).innerHTML = order.map(f => {
+    const [se, ra, di, total] = board[f];
+    return '<div class="stack"><div class="lab">' + NAME[f] + '</div><div class="trk">' +
+      '<div class="seg" style="width:' + (se/80*100) + '%;background:#1f5fdb"></div>' +
+      '<div class="seg" style="width:' + (ra/80*100) + '%;background:#e0892e"></div>' +
+      '<div class="seg" style="width:' + (di/80*100) + '%;background:#0f9d8a"></div>' +
+      '</div><div class="val">' + total.toFixed(1) + '</div></div>';
+  }).join("");
+}
+renderScores("sc-measured", SCORES.measured);
+renderScores("sc-proj", SCORES.proj);
+</script>
+</body>
+</html>

docs/rules/bench-methodology.md

@@ -0,0 +1,121 @@
+# Bench / Methodology - isolation, repetitions, timeouts, honest reporting
+
+**Purpose + scope.** The measurement discipline for the benchmark matrix
+(Q1-Q7 x 5 formats, spec [08](../specs/08_benchmark_queries.md) /
+[09](../specs/09_benchmark_execution.md)) and for the extrapolation and
+reporting built on it (specs 10-11). Binding for `bench_runner.py`, every
+query implementation, `extrapolate.py`, and `report.py`.
+
+The product of this project IS these measurements; a sloppy protocol
+invalidates everything downstream.
+
+---
+
+## 1. Cell isolation - one subprocess per measurement
+
+- Every benchmark cell (format x query x run) executes in its OWN
+  subprocess, so peak RSS, CPU time, and IO counters attribute to exactly
+  that cell.
+- The runner samples the child with `psutil` every 50 ms for peak RSS;
+  CPU time is user+sys at exit; read bytes from `io_counters`.
+- Nothing else runs in the child: no logging setup beyond minimum, no
+  imports the query does not need - the measurement must not pay for the
+  harness.
+
+## 2. Repetition protocol
+
+- Per cell: **1 warm-up run (discarded) + 3 measured runs.** Report the
+  median in `results_median.csv`; keep min/max in the raw data.
+- **Randomize the cell execution order** and record the actual order in
+  the raw results - fixed order lets cache effects masquerade as format
+  differences.
+- SQLite additionally reports run 1 (cold) and run 3 (warm) separately
+  (spec 08).
+- Never average TIMEOUT/ERROR runs into medians; a cell with a failed run
+  reports the failure, not a prettier subset.
+
+## 3. Timeouts and failures are data
+
+- Hard per-run timeout: **30 minutes.** On expiry the runner records
+  `TIMEOUT` for that cell and CONTINUES the matrix (an expected outcome:
+  RDF Q5). Never retry a timeout hoping for a better number.
+- A crashed cell records `ERROR` the same way (see
+  [code-error-handling.md](code-error-handling.md) pattern C).
+- In extrapolation and scoring, TIMEOUT/FAIL project to score 0 - they
+  are never silently dropped from tables or charts (spec 10/11 flags:
+  projection > 1 h = `IMPRACTICAL`, > 24 h = `FAIL`).
+
+## 4. Result correctness gates the measurement
+
+- Every measured run validates its result rows against the canonical
+  checksum (sorted rows, float tolerance 1e-9) and records `checksum_ok`
+  (see [test-pipeline-validation.md](test-pipeline-validation.md)).
+- A fast run with `checksum_ok = false` is an INVALID measurement - it
+  must never enter medians, extrapolation, or the report.
+- **Q5 reads raw `friction_cycles` in every format** - summary tables and
+  materialized views are forbidden for it; verify PostgreSQL did not
+  satisfy it from `track_summary`
+  (`EXPLAIN (ANALYZE, BUFFERS)` is recorded for PostgreSQL runs anyway).
+
+## 5. Fairness across formats
+
+- Identical result-set contract for all 5 implementations of each query;
+  no format answers a simplified question.
+- Each format uses its idiomatic strength honestly: CSV Q1 reads the one
+  target file by path; JSON Q1 loads the one coupon file; SQL uses its
+  indexes. What is forbidden is capping or sampling (`LIMIT`, partial
+  scans) that changes the answer.
+- The pandas CSV variant is measured as a SEPARATE variant, never mixed
+  into the plain-CSV numbers (spec 08).
+- Storage sizes compare like with like: for every format record both the
+  working footprint and the gzip archival footprint in
+  `storage_sizes.csv` (spec 09).
+
+## 6. Environment honesty - this machine, stated limits
+
+- Record the hardware and software environment (CPU model, cores, RAM,
+  disk type, OS, engine versions) once per benchmark session into the
+  raw results; the report cites it.
+- **Cold-cache protocol on this host:** the spec's page-cache drop
+  (`/proc/sys/vm/drop_caches`) is Linux-only; this project runs on
+  Windows Server 2025, where it is unavailable. The cold-cache matrix
+  pass is therefore SKIPPED and marked as such in results and report -
+  never approximated by a fake (e.g. file re-copy tricks) without a
+  rules update describing the method.
+- **PostgreSQL is remote on this setup** (dedicated DB host on the LAN),
+  so spec 09's backend-PID sampling via `pg_stat_activity` + psutil is
+  impossible - psutil cannot attribute a remote process. Instead the
+  runner records (a) client-side child metrics exactly like every other
+  format (wall times therefore include LAN round-trips - stated in the
+  report), (b) best-effort Prometheus / node_exporter host CPU/RAM
+  windows per PG cell and for the whole session (`TRIBO_PROM_URL`), and
+  (c) best-effort `pg_stat_statements` per-query session deltas
+  (server-side execution time and buffer traffic).
+- Quiesce the machine during measured runs (no parallel pipeline tasks,
+  no background loads); the runner runs cells sequentially by design.
+
+## 7. Extrapolation discipline (task 10)
+
+- Scaling laws per access pattern are those fixed in spec 10 (O(1),
+  O(log n + k), O(k log n), O(n), O(n / min(cores, partitions)));
+  constants calibrate on the measured 150 MB point.
+- Every projected number in the report is labeled as projected, with its
+  scaling assumption; measured and projected values never mix in one
+  column unlabeled.
+- Hardware cost figures come from `hw_prices.yaml`
+  (see [code-config-yaml.md](code-config-yaml.md)), never inlined.
+
+## 8. Anti-patterns
+
+- **Running cells in-process** - RSS/CPU attribution becomes fiction.
+- **Fixed execution order** - cache warmth leaks between formats.
+- **Retrying or dropping slow runs** - the slow run IS the result.
+- **Averaging over a TIMEOUT** or reporting medians of fewer than 3 valid
+  runs without flagging it.
+- **A "fast" result that fails the checksum** entering any downstream
+  table.
+- **Tuning one format's query while leaving others naive** - fairness is
+  per-scenario, not per-format-champion.
+- **Faking cold-cache on Windows** without a documented, rules-approved
+  method.
+- **Presenting projected numbers as measured** in tables or charts.

docs/rules/build-pipeline-tasks.md

@@ -0,0 +1,109 @@
+# Build / Pipeline tasks - the ./out tree, .done markers, task order
+
+**Purpose + scope.** How the 11 pipeline tasks compose: where artifacts
+live, how completion is signaled, what order tasks run in, and what git
+does and does not track. Binding for every task implementation and for
+anyone running the pipeline.
+
+---
+
+## 1. The artifact tree - everything under `./out/`
+
+All generated artifacts live under `./out/` relative to the repo root
+(spec 00_PLAN):
+
+```
+out/
+  config/    lab_config.yaml, hw_prices.yaml
+  csv/       the canonical raw corpus + MANIFEST.csv
+  json/      full/ (per-coupon JSON-LD), hybrid/dataset.jsonld
+  sqlite/    tribo.db
+  pg/        tribo.dump (pg_dump -Fc; the live DB is in PostgreSQL)
+  rdf/       dataset.nt.gz, dataset.ttl, oxigraph_store/
+  bench/     queries/, expected/, results_raw.csv, results_median.csv,
+             storage_sizes.csv, extrapolation.csv, hardware_sizing.csv
+  report/    REPORT.md, charts/, tables/, process_flow.md, diagrams/
+  .done/     completion markers, one per task
+```
+
+- A task writes ONLY under its own output paths; it treats its inputs as
+  read-only.
+- Tasks communicate exclusively through these artifacts - never through
+  shared in-process state
+  (see [code-python-style.md](code-python-style.md)).
+- **The bulk of `./out/` is git-ignored.** Sources of truth are the
+  specs, the rules, and the code; artifacts are reproducible from them
+  ([data-determinism.md](data-determinism.md)). A curated results subset
+  IS committed (owner decision, 2026-07-11): `out/report/` (REPORT.md,
+  charts, tables, diagrams), `out/config/` (lab_config.yaml,
+  hw_prices.yaml), and the `out/.done/` markers - exactly the exceptions
+  listed in `.gitignore`. Never commit the bulk artifacts (corpus,
+  databases, stores, archives, raw benchmark data) "for convenience".
+
+## 2. Completion markers
+
+- Each task NN writes `./out/.done/<NN>.ok` after - and only after - its
+  outputs are complete and validated. The marker body carries the
+  headline counts (rows, bytes, triples) downstream tasks check
+  (see [test-pipeline-validation.md](test-pipeline-validation.md)).
+- A task verifies its dependencies' markers before starting and aborts
+  loudly if one is missing.
+- Re-running a task: delete its own marker first, regenerate outputs,
+  rewrite the marker. If its outputs changed, downstream markers are now
+  stale - delete them too (the dependency graph below says which).
+
+## 3. Task graph and execution order
+
+```
+01 -> 02
+01 -> 03 -> (04 | 05 | 06 | 07) -> 08 -> 09 -> 10 -> 11
+                                          09 ------> 11
+```
+
+- Sequential order: 01, 02, 03, then 04-07 (independent, may run in
+  parallel), 08, 09, 10, 11.
+- 02 (process diagrams) needs only 01 and can run any time after it.
+- Task 09 (benchmark execution) must run WITHOUT 04-07 style parallel
+  work in the background - measurements need a quiet machine
+  ([bench-methodology.md](bench-methodology.md) section 6).
+
+## 4. `storage_sizes.csv` - the shared append contract
+
+- Location: `./out/bench/storage_sizes.csv`; columns:
+  `format, variant, bytes`.
+- Tasks 04-07 APPEND their format's rows (e.g. `json,full,...`,
+  `json,hybrid,...`, `sqlite,db,...`, `pg,live,...`, `pg,dump,...`,
+  `rdf,nt_gz,...`, `rdf,store,...`); task 09 appends the gzip archival
+  variants and consolidates.
+- Appends are idempotent per (format, variant): a re-run REPLACES its own
+  prior rows, never duplicates them and never touches other formats'
+  rows.
+- RDF load wall-time and similar per-format load metrics recorded per
+  spec 07 go into the task's `.done` marker and the report, not into
+  extra ad-hoc files.
+
+## 5. Runs are resumable, not restartable-from-zero
+
+- The marker system exists so a failed pipeline resumes at the failed
+  task; do not delete the whole `./out/` tree to retry one task.
+- Conversely, after a deliberate corpus regeneration (03), ALL downstream
+  markers and artifacts (04-11) are stale and must be removed - a mixed
+  tree of old and new artifacts is worse than an empty one
+  ([meta-no-assumptions.md](meta-no-assumptions.md) applies before any
+  such deletion).
+
+## 6. Anti-patterns
+
+- **Writing an artifact outside `./out/`** or a task writing into another
+  task's output directory.
+- **Committing bulk `./out/` content to git** (corpus, databases, stores,
+  archives, raw benchmark data) - only the curated subset in `.gitignore`
+  (`report/`, `config/`, `.done/`) is committed.
+- **A marker written on partial success** - see
+  [code-error-handling.md](code-error-handling.md).
+- **Running benchmarks while converters are still running.**
+- **Duplicate rows in `storage_sizes.csv`** after re-runs - replace your
+  own rows.
+- **Hand-editing generated artifacts** (a "quick fix" inside
+  `MANIFEST.csv` or a result CSV) - regenerate them through the owning
+  task instead.

docs/rules/code-comments-policy.md

@@ -0,0 +1,61 @@
+# Code comments - write WHY, not WHAT
+
+**Binding** for all source files: Python, SQL, SPARQL, PowerShell, YAML, shell.
+
+## Rule
+
+Default to **no comment**. Add one only when the *why* is non-obvious from
+the code. If removing the comment wouldn't confuse a future reader, delete it.
+
+A comment earns its place when it captures something the code can't:
+hidden constraint, ordering invariant, workaround for a named bug, vendor
+quirk, cross-file invariant the type system can't enforce.
+
+## Delete on sight
+
+```python
+# Read the manifest file
+manifest = read_manifest(csv_root)          # function name says it
+```
+
+```python
+# Increment counter
+counter += 1                                # pure restatement
+```
+
+```python
+# Added for the benchmark rework (see task 09)
+def run_cell(...):                          # belongs in commit message, rots in code
+```
+
+```python
+# TODO: refactor later
+                                            # no owner, no trigger - decoration
+```
+
+## Keep
+
+```python
+# Q5 must scan raw friction_cycles - track_summary is forbidden for this
+# query in every format (spec 08); do not "optimize" this into the summary.
+```
+Cross-file invariant a refactor must preserve.
+
+```python
+# Column order matches MANIFEST.csv writer in generate_data.py - keep in sync.
+```
+Cross-file invariant the type system can't enforce.
+
+```sql
+-- BRIN, not b-tree: cycle values are physically ordered within each
+-- partition, so BRIN gives the same pruning at ~1% of the index size.
+```
+Non-obvious engineering decision with a reason.
+
+## Length
+
+Inline: one line. Block: 1-4 lines. If you need more than a paragraph,
+the right home is a rule or spec doc - link it from a one-line comment.
+
+Stale comments are worse than missing ones. When you touch a file, fix or
+delete comments whose *why* no longer holds.

docs/rules/code-config-yaml.md

@@ -0,0 +1,94 @@
+# Code / Config - all tunables live in YAML, no magic numbers, no .env
+
+**Purpose + scope.** Governs where every configurable value lives. Binding
+for all Python code in this repository. Any new configurable parameter MUST
+follow this rule when deciding where it goes.
+
+The goal: one authoritative source per parameter, so regenerating any
+artifact with the same config is byte-identical, and changing a parameter
+never requires hunting constants across scripts.
+
+---
+
+## 1. The two YAML configs
+
+| File | Owns | Written by |
+|---|---|---|
+| `./out/config/lab_config.yaml` | The laboratory model: volumes (batches / wafers / coupons / tracks / cycles), the deposition matrix, physical-model coefficients, nominal test conditions, **the random seed 20260711** | Task 01, from the spec; regenerated only deliberately |
+| `./out/config/hw_prices.yaml` | Hardware cost parameters for extrapolation (RAM $/GB, NVMe $/TB, server base prices) | Task 10 seeds editable defaults; the operator may edit |
+
+Rules:
+
+- **Scripts READ these files; they never duplicate values from them.**
+  Task 03 reads volumes, coefficients, and the seed from
+  `lab_config.yaml` - it does not hardcode 588 coupons or seed 20260711.
+  Task 10 reads prices from `hw_prices.yaml` - it does not inline dollar
+  figures.
+- A parameter needed by more than one task lives in the YAML, never
+  copy-pasted between scripts.
+- New tunables go into the appropriate YAML with a self-describing key;
+  if neither file fits, propose a new config file in a docs/rules update
+  first.
+
+## 2. Spec-fixed constants - named, sourced, at the top
+
+Values fixed by the specs that are NOT operator-tunable (the 120-200 MB
+size gate, the 30-minute cell timeout, the 50k-row transaction batch, the
+50 ms RSS sampling interval) are module-level named constants at the top of
+the owning script, each with a one-line comment naming the spec:
+
+```python
+SIZE_GATE_MB = (120, 200)      # abort bounds, spec 03
+CELL_TIMEOUT_S = 30 * 60       # hard per-run timeout, spec 09
+```
+
+A bare number inline in the logic (`if size > 209715200`) is forbidden.
+
+## 3. The PostgreSQL DSN - one env var, never hardcoded, never logged
+
+The only external resource is the local PostgreSQL 16 instance. Its DSN:
+
+- Comes from the environment variable **`TRIBO_PG_DSN`**; default when
+  unset: `postgresql://postgres@localhost:5432/tribo`.
+- Is read in ONE shared helper (e.g. `common/pg.py: get_dsn()`); scripts
+  call the helper, never `os.environ` directly.
+- Is never hardcoded in a script, never committed in a config file, and
+  never logged (a DSN may embed a password - see
+  [obs-python-logging.md](obs-python-logging.md)).
+
+There are no other secrets in this project. There is no `.env` file; do
+not introduce one.
+
+## 4. CLI flags - per-run choices only
+
+Command-line flags are for choices that vary per invocation, not for
+model parameters: `--log-level`, `--dry-run`, an output-root override, a
+`--limit` for smoke runs. A physical coefficient or a volume as a CLI flag
+is forbidden - it would silently fork the config out of `lab_config.yaml`.
+
+## 5. Anti-patterns
+
+- **Hardcoding a value that exists in `lab_config.yaml`** (seed, volumes,
+  coefficients) - the YAML is the single source; drift here breaks
+  reproducibility silently.
+- **Adding a `.env` file or reading `os.environ` scattered through
+  scripts** - the DSN goes through the one shared helper; nothing else
+  lives in the environment.
+- **Inline magic numbers** for spec-fixed gates and timeouts - name them
+  at the top with the spec reference.
+- **Dollar prices inlined in `extrapolate.py`** - they live in
+  `hw_prices.yaml` so the operator can re-cost without touching code.
+- **A model parameter as a CLI flag** - per-run flags are for run
+  mechanics, not for the lab model.
+- **Committing `./out/config/` generated files to git as if they were
+  sources** - the YAML under `./out/` is a generated artifact; its source
+  of truth is spec 01 (see
+  [build-pipeline-tasks.md](build-pipeline-tasks.md)).
+
+## 6. Related documents
+
+- [data-determinism.md](data-determinism.md) - the seed contract.
+- [build-pipeline-tasks.md](build-pipeline-tasks.md) - the `./out/` tree
+  and what is git-ignored.
+- [obs-python-logging.md](obs-python-logging.md) - `--log-level` flag; DSN
+  never logged.

docs/rules/code-data-formatting.md

@@ -0,0 +1,129 @@
+# Code / Data / Formatting - ASCII-only typographic characters in source files
+
+**Binding scope.** Strings that end up in any of the following MUST use ASCII
+typographic characters; never their Unicode "smart" equivalents:
+
+- Python source, SQL and SPARQL query files
+- Generated data and artifacts written to disk (CSV, JSON-LD, N-Triples,
+  Turtle, MANIFEST and benchmark result CSVs, log messages)
+- Configuration files (`.yaml`, `.json`, `.toml`)
+- PowerShell / shell scripts (any file processed by Windows tooling)
+- Markdown documentation (`docs/`, `README.md`, `CLAUDE.md`) - the specs
+  were normalized to ASCII typography on 2026-07-11 and stay that way
+
+**Long dashes are an absolute exception, no carve-outs.** Em-dash (U+2014)
+and en-dash (U+2013) MUST be replaced with ASCII hyphen-minus (`-`) in
+**every** file in this repository, including markdown documentation, code
+comments, and report prose. Reason: long dashes are visually
+indistinguishable from a hyphen in many monospace fonts, break diffs / grep /
+IDE search, and round-trip badly through Windows tooling chains (section 2).
+
+---
+
+## 1. The substitution table
+
+| Unicode char | Codepoint | Name | ASCII replacement |
+|---|---|---|---|
+| em-dash | U+2014 | em-dash | `-` (single hyphen-minus) |
+| en-dash | U+2013 | en-dash | `-` (single hyphen-minus) |
+| horizontal bar | U+2015 | horizontal bar | `-` |
+| minus sign | U+2212 | minus sign | `-` |
+| ellipsis | U+2026 | horizontal ellipsis | `...` (three ASCII dots) |
+| left double quote | U+201C | left double quotation mark | `"` |
+| right double quote | U+201D | right double quotation mark | `"` |
+| left single quote | U+2018 | left single quotation mark | `'` |
+| right single quote | U+2019 | right single quotation mark | `'` |
+| left guillemet | U+00AB | left guillemet | `"` |
+| right guillemet | U+00BB | right guillemet | `"` |
+| nbsp | U+00A0 | non-breaking space | regular space |
+
+**Single hyphen, not double.** When replacing an em-dash that separates two
+clauses, use a single `-` surrounded by spaces (`text - clause`), NOT `--`.
+
+**Ranges use a single hyphen:** `Q1-Q7`, `120-200 MB`, `0.3-1.1 um` - never
+an en-dash.
+
+## 2. Why - documented failure modes
+
+1. **PowerShell 5.1 reads no-BOM files as Windows-1252.** Non-ASCII UTF-8
+   bytes become mojibake and break string-literal parsing. This project is
+   developed on Windows; any script or config touched by PowerShell is
+   exposed.
+
+2. **Windows -> git-bash -> tar / packaging round-trip.** At any step an
+   unset locale or wrong encoding assumption produces mojibake. A UTF-8 byte
+   sequence for U+2014 interpreted under a non-UTF-8 locale renders as a
+   `?` replacement character in the packaged output.
+
+3. **Copy-paste from Word / Google Docs / Notion / web articles.** These
+   tools auto-substitute typography (`"` -> smart quotes, `'` -> smart
+   apostrophe, `--` -> em-dash, `...` -> ellipsis) silently. The
+   substitution survives copy into any editor and lands in source files
+   unnoticed. The original specs arrived with en/em dashes this way.
+
+4. **Data files must diff and grep cleanly.** The CSV corpus, MANIFEST, and
+   benchmark results are compared by checksum and processed by streaming
+   parsers; a stray typographic character breaks byte-level reproducibility.
+
+## 3. Where the rule does NOT apply
+
+- **Scientific and unit notation is content, not typography.** Characters
+  such as `µ` (micro), `±`, `°C`, `×`, `Ø`, and Greek letters (`τ`, `σ`) in
+  formulas inside the specs are meaningful scientific notation and are kept.
+  Note the boundary: in **column names and code identifiers** the units are
+  always plain ASCII (`_um`, `_nm`, `_GPa` - see
+  [data-naming-units.md](data-naming-units.md)); the scientific characters
+  live only in explanatory prose.
+- **International language strings** (Cyrillic, CJK, etc.) are meaningful
+  content, not typographic substitutions. UTF-8 throughout is the right
+  answer. (Repository artifacts are English; this matters only for quoted
+  material.)
+
+## 4. Detection / enforcement
+
+There is no automated check for this rule yet - **not in CI yet**. The rule is
+enforced by review and by the editing discipline in this file. The intent of a
+future detector is documented here so it can be built consistently:
+
+- **What it flags.** Any occurrence of the Unicode codepoints in the section 1
+  substitution table inside in-scope files. The section 3 exemptions
+  (scientific notation, international content) are the only carve-outs.
+- **Exit behavior.** Non-zero exit when any flagged codepoint is found under
+  the scan roots, so it can later gate a commit hook or a build step.
+- **Tooling.** When this detector is created it MUST be a PowerShell script
+  (this is a Windows project) and live under `tools/`.
+
+## 5. Examples
+
+### Good
+
+```python
+# safe through every Windows tooling step
+print("Corpus size 152.3 MB - within the 120-200 MB gate.")
+label = "steady-state COF, cycles 500-1000"
+```
+
+### Bad
+
+```python
+# em-dash literal (U+2014) - mojibake risk on PowerShell read/write
+print("Corpus size 152.3 MB [U+2014] within the gate.")
+
+# en-dash range (U+2013) - breaks grep for "Q1-Q7"
+label = "queries Q1[U+2013]Q7"
+```
+
+## 6. Anti-patterns
+
+- **Copy-pasting strings from a word processor or web page** into source /
+  specs / config without normalization.
+- **Trusting tar / packaging to "preserve as-is".** They preserve bytes; the
+  *encoding interpretation* is what gets lost when locale or encoding
+  declaration is wrong on either end.
+- **Assuming the source file encoding alone is enough.** The source file may
+  be fine - the failure is at downstream consumers (PowerShell read, build
+  write, OS locale).
+- **Using `--` instead of `-` "for emphasis".** Single ASCII hyphen
+  surrounded by spaces is the project's chosen separator.
+- **Putting `µ` or `°` into a column name or identifier** - units in names
+  are ASCII (`_um`, `_C`); see [data-naming-units.md](data-naming-units.md).

docs/rules/code-error-handling.md

@@ -0,0 +1,181 @@
+# Code / Error handling - never swallow, always surface, fail loud
+
+**Purpose + scope.** Governs every `try:` / `except:` / `raise` site in
+Python code in this repository (data generation, format converters, benchmark
+runner, extrapolation, reporting, helper tools). Binding for all contributors.
+
+The pipeline's core failure principle: **a wrong artifact is worse than no
+artifact.** Every task's output feeds downstream tasks; silently degraded
+data poisons every result built on it. When in doubt, abort with a clear
+message and a non-zero exit code.
+
+---
+
+## 1. The rule
+
+1. **Never `except: pass`.** A bare swallow is always a bug. Forbidden
+   absolutely.
+2. **Never `except Exception: logger.warning("...") + continue`** for
+   anything that was not an EXPECTED, documented, transient condition. If
+   you swallow and continue, you must be able to state why the program can
+   correctly proceed.
+3. **Always `logger.error(..., exc_info=True)` in `except` blocks.**
+   `exc_info=True` captures the traceback into the log record (see
+   [obs-python-logging.md](obs-python-logging.md)). Without it the original
+   exception is lost.
+4. **Re-raise unless you have a defined recovery path.** If the caller
+   cannot continue meaningfully without the failed operation, re-raise
+   after logging.
+5. **Surface to the right channel:**
+   - **CLI script (every pipeline task):** print the error to stderr and
+     exit non-zero. A task that failed MUST NOT write its
+     `./out/.done/<task>.ok` marker (see
+     [build-pipeline-tasks.md](build-pipeline-tasks.md)).
+   - **Validation gate (row counts, checksums, size bounds):** a failed
+     gate is an abort, not a warning. Example: task 03 aborts when the
+     generated tree is outside 120-200 MB; converters abort when row counts
+     do not match `MANIFEST.csv`.
+   - **Library / helper function:** re-raise; let the caller decide.
+6. **Defensive "this cannot happen" branches must surface, not silently
+   `return`.** If you reach a defensive branch, log at ERROR with the
+   branch name and the inputs that got you there, then raise.
+
+## 2. The exception ladder - the only three sanctioned swallow patterns
+
+Code may swallow ONLY in these three documented patterns. Anything else
+re-raises after logging.
+
+### Pattern A: documented, recoverable transient (retry)
+
+```python
+for attempt in range(MAX_RETRIES):
+	try:
+		return do_thing()
+	except (psycopg.OperationalError, TimeoutError) as exc:
+		logger.warning("attempt %d/%d failed: %s", attempt + 1, MAX_RETRIES, exc)
+		time.sleep(backoff(attempt))
+# All retries exhausted - surface.
+logger.error("do_thing failed after %d retries", MAX_RETRIES)
+raise
+```
+
+Justified: the per-attempt warning is acceptable; the final failure
+re-raises.
+
+### Pattern B: optional best-effort cleanup
+
+```python
+def _close_conn(conn) -> None:
+	try:
+		conn.close()
+	except Exception:
+		# Cleanup is best-effort; the operation already finished, log and continue.
+		logger.warning("closing connection failed", exc_info=True)
+```
+
+Justified: the comment names why continuing is correct; the traceback is
+still captured.
+
+### Pattern C: benchmark-cell isolation - record and continue the matrix
+
+```python
+try:
+	result = run_cell(fmt, query, timeout_s=30 * 60)
+except CellTimeout:
+	logger.error("cell %s/%s timed out", fmt, query)
+	write_result_row(fmt, query, status="TIMEOUT")   # recorded, not hidden
+except Exception:
+	logger.error("cell %s/%s failed", fmt, query, exc_info=True)
+	write_result_row(fmt, query, status="ERROR")
+```
+
+Justified: spec 09 requires the benchmark matrix to survive a failing cell
+(RDF Q5 is expected to time out). The failure is RECORDED as data
+(`TIMEOUT` / `ERROR` in `results_raw.csv`), never silently dropped, and the
+run continues with the next cell. This pattern applies ONLY to the benchmark
+runner's per-cell boundary - a failure in the runner itself (setup, result
+writing) still aborts.
+
+## 3. Loading into SQLite / PostgreSQL - all-or-nothing per table
+
+A converter that half-loads a table leaves a database that LOOKS complete
+and silently skews every benchmark run against it.
+
+1. Wrap each table's bulk load in a transaction; on any error roll back
+   and abort the converter with a non-zero exit.
+2. After loading, validate row counts against `MANIFEST.csv` and fail loud
+   on any mismatch (see
+   [test-pipeline-validation.md](test-pipeline-validation.md)).
+3. Never write the task's `.done` marker after a partial load.
+
+## 4. Anti-patterns
+
+- **`except: pass`** - forbidden absolutely.
+- **`except Exception: logger.warning("could not X")` + continue** -
+  forbidden unless the function documents that "could not X" is a
+  recoverable expected outcome. If it is, phrase it "X skipped (expected
+  condition Y)" so a reader does not grep "could not" hunting for bugs.
+- **Catching `BaseException`** - forbidden (swallows `KeyboardInterrupt`,
+  `SystemExit`). Use `Exception`.
+- **`logger.error("X")` without `exc_info=True` in an `except` block** -
+  forbidden. The traceback IS the diagnostic.
+- **`logger.error(str(exc))`** - forbidden; redundant with `exc_info=True`
+  and loses type and stack.
+- **Returning `None` to signal failure** from a function whose contract is
+  "return T" - forbidden. Either raise, or return `T | None` with
+  documented semantics the caller checks.
+- **`raise Exception("...")`** - forbidden. Use a specific class:
+  `ValueError` / `RuntimeError` / `KeyError` for trivial cases, a custom
+  domain exception subclass (e.g. `ManifestMismatchError`) for domain
+  conditions.
+- **`raise` chained without context** - inside an `except`, prefer
+  `raise NewError(...) from exc` so the original cause stays visible in the
+  chained traceback. A bare `raise` that re-raises the same exception is
+  fine; `raise SomethingElse(...)` without `from exc` is forbidden.
+- **Writing a `.done` marker in a `finally` block** - the marker means
+  "completed successfully", never "attempted".
+- **Clamping / repairing out-of-range generated values silently** - a
+  physical-model bug must surface, not be masked by a `max(0, ...)` added
+  "to be safe". Deliberate clamps defined by the spec (e.g. COF floor 0.12)
+  are model logic, not error handling.
+
+## 5. Defensive branches
+
+```python
+# Good:
+def open_store(fmt: str) -> Store:
+	if fmt == "sqlite":
+		return SqliteStore()
+	if fmt == "postgresql":
+		return PgStore()
+	# Defensive: a new format added without updating this factory lands here.
+	logger.error("open_store: unknown format=%r", fmt)
+	raise ValueError(f"unknown storage format: {fmt!r}")
+```
+
+The silent-`return None` form lets the unknown value propagate and breaks
+downstream code with a confusing `AttributeError` on the wrong line.
+
+## 6. Logging vs surfacing - the decision
+
+For a caught exception, ask in order:
+
+1. **Will the pipeline produce a wrong artifact if I continue?** Re-raise /
+   abort.
+2. **Is this expected and recoverable in the caller's contract?** Log at
+   WARNING with `exc_info=True`, return a sentinel the caller checks.
+3. **Is this an isolated benchmark cell (pattern C)?** Log at ERROR, record
+   the status as data, continue the matrix.
+4. **Is the process going down anyway?** Log at CRITICAL with
+   `exc_info=True`, then `sys.exit(1)` or let it propagate.
+
+## 7. Related documents
+
+- [obs-python-logging.md](obs-python-logging.md) - logging API and severity
+  levels.
+- [test-pipeline-validation.md](test-pipeline-validation.md) - the
+  validation gates that turn silent corruption into loud failures.
+- [build-pipeline-tasks.md](build-pipeline-tasks.md) - `.done` marker
+  semantics.
+- [bench-methodology.md](bench-methodology.md) - the TIMEOUT contract for
+  benchmark cells.

docs/rules/code-python-style.md

@@ -0,0 +1,93 @@
+# Code / Python style - tabs, 3.11+, closed dependency list, streaming
+
+**Purpose + scope.** Baseline style and dependency policy for every Python
+file in this repository (pipeline tasks 01-11, query implementations,
+helper tools). Binding for all contributors.
+
+---
+
+## 1. Language and formatting
+
+- **Python 3.11+.** Use modern stdlib freely (`pathlib`, `dataclasses`,
+  `tomllib` if needed); no compatibility shims for older versions.
+- **Indentation: TABS.** Set by the project owner and repeated in the
+  specs (00_PLAN, 03, 09). Every Python file in this repository indents
+  with tab characters, never spaces. Configure the editor per file type;
+  do not "fix" tab-indented files to spaces.
+- `from __future__ import annotations` at the top of every module.
+- Type hints on public function signatures; internal helpers at author's
+  discretion.
+- English for all identifiers, docstrings, and comments (see
+  [meta-language-style.md](meta-language-style.md) - Russian is the
+  conversation language, not the code language).
+- Comments follow [code-comments-policy.md](code-comments-policy.md):
+  default to none, keep only non-obvious WHY.
+
+## 2. Dependencies - a closed list
+
+The stdlib is the default. The ONLY permitted third-party packages:
+
+| Package | Used by | For |
+|---|---|---|
+| `numpy` | 03 | vectorized generation, seeded RNG |
+| `PyYAML` | 01, 03, 10, 11 | reading/writing the YAML configs |
+| `rdflib` | 04, 07 | JSON-LD validation, RDF serialization |
+| `oxigraph` | 07, 09 | embedded queryable triplestore |
+| `ijson` | 08, 09 | streaming JSON parsing |
+| `psycopg` (v3) | 06, 08, 09 | PostgreSQL COPY and queries |
+| `psutil` | 04-07, 09 | RSS / CPU / IO measurement; converters record their own peak RSS / CPU time into their `.done` markers for the task 10 sizing model |
+| `matplotlib` | 11 | report charts |
+| `pandas` | 08 (optional) | ONLY as the separately-measured second CSV query variant; never in the pipeline itself |
+
+Adding any other package requires updating this table first. In
+particular forbidden: ORMs, logging frameworks
+([obs-python-logging.md](obs-python-logging.md)), CLI frameworks
+(argparse suffices), `requests`/HTTP clients (nothing to call).
+
+Task 03 (data generation) is stricter by spec: **stdlib + numpy only.**
+
+## 3. Script shape - every task is a self-contained CLI
+
+- One entry script per pipeline task, runnable as
+  `python <script>.py [flags]`.
+- `argparse` with `--log-level` (see
+  [obs-python-logging.md](obs-python-logging.md)) and task-appropriate
+  per-run flags ([code-config-yaml.md](code-config-yaml.md) section 4).
+- A module docstring stating the task number, what it produces, and its
+  spec file.
+- `main() -> int` returning the exit code; `sys.exit(main())` under
+  `if __name__ == "__main__":`.
+- Exit code 0 only on full success (outputs written, validation passed,
+  marker written).
+
+## 4. Streaming discipline - the corpus never fits in memory by design
+
+The generated corpus is ~150 MB and its scaled-up target is 600 GB; code
+must behave as if the data never fits in RAM:
+
+- **Write generated rows as you produce them**; never accumulate the full
+  cycle/loop dataset in a list (spec 03: "stream rows; never hold the
+  full cycle dataset in memory").
+- **Read CSV/JSON with iterators** (`csv.reader` row-by-row, `ijson`
+  events); never `read()` a bulk file whole or `json.load` the full
+  corpus.
+- Batch database work (`executemany` / `COPY` in chunks), keeping only
+  the current batch in memory.
+- Bounded aggregation state is fine (per-track summaries, counters);
+  unbounded accumulation of raw rows is the thing being forbidden.
+
+## 5. Anti-patterns
+
+- **Spaces for indentation** in Python files - the project standard is
+  tabs.
+- **An unlisted third-party import** - extend the section 2 table first
+  or use the stdlib.
+- **`pd.read_csv` in pipeline code** - pandas exists only as the
+  separately-measured Q-variant in spec 08.
+- **Loading a bulk artifact whole** (`json.load` on the full corpus,
+  `readlines()` on a cycles CSV) - stream it.
+- **A task importable only from a specific CWD** - resolve paths from a
+  repo-root anchor, not the current directory.
+- **Two tasks sharing mutable module state** - tasks communicate through
+  `./out/` artifacts only (see
+  [build-pipeline-tasks.md](build-pipeline-tasks.md)).

docs/rules/data-determinism.md

@@ -0,0 +1,85 @@
+# Data / Determinism - one seed, byte-identical regeneration
+
+**Purpose + scope.** The reproducibility contract for all generated data.
+Binding for task 03 (CSV generation) and for every piece of code that
+produces values which end up inside the corpus or its conversions.
+
+The entire evaluation rests on this: benchmark results are only comparable
+because every format holds EXACTLY the same data, and the study is only
+reviewable because anyone can regenerate that data bit-for-bit.
+
+---
+
+## 1. The seed
+
+- **The global seed is `20260711`.** It lives in `lab_config.yaml`
+  (written by task 01) and is READ from there - never hardcoded in a
+  generation script (see [code-config-yaml.md](code-config-yaml.md)).
+- Changing the seed is a project-level decision, never a side effect. It
+  invalidates the whole `./out/` tree, every checksum, and every measured
+  result (see [meta-no-assumptions.md](meta-no-assumptions.md)).
+
+## 2. Sanctioned randomness - seeded generators only
+
+- The ONLY random sources are `random.Random(seed)` and
+  `numpy.random.default_rng(seed)`, constructed explicitly from the
+  config seed.
+- Derive per-entity generators deterministically when parallel or
+  out-of-order generation needs them (e.g.
+  `default_rng([seed, batch_index, coupon_index])`) - never from time,
+  never from a global unseeded state.
+- Forbidden anywhere in generation code: module-level `random.*` calls,
+  `numpy.random.*` legacy global functions, `os.urandom`, `uuid.uuid4`,
+  `hash()`-dependent ordering (Python string hashing is salted per
+  process).
+
+## 3. No wall-clock in the data
+
+- Every timestamp inside the corpus is SIMULATED - computed from the lab
+  schedule model - never `datetime.now()` / `time.time()`.
+- Real wall-clock time may appear only in logs and in measurement records
+  (benchmark wall times in `./out/bench/` are measurements, not corpus
+  data).
+
+## 4. Deterministic everything else
+
+Randomness is not the only nondeterminism; all of these are binding:
+
+- **Iteration order.** Iterate in a defined order (sorted paths, explicit
+  index loops). Never rely on `os.listdir` / `glob` / `set` / dict-of-set
+  ordering for anything that affects output. Sort directory listings
+  before use.
+- **Float formatting.** Format floats with an explicit spec (e.g.
+  `f"{x:.6g}"`), chosen once per column family; repr drift between runs
+  or Python versions must not change file bytes.
+- **Line endings and encoding.** Generated files are UTF-8 with `\n`
+  newlines, on every platform (`open(..., newline="")` for csv writers).
+- **Parallelism.** If generation is parallelized, each worker's output is
+  deterministic (derived generators, fixed partitioning) and the merge
+  order is fixed.
+
+## 5. The proof - MANIFEST sha256
+
+- `./out/csv/MANIFEST.csv` records sha256 per generated file (spec 03).
+  **Regenerating with the same `lab_config.yaml` MUST reproduce every
+  sha256 exactly.** This is the acceptance test for this whole rule -
+  run generation twice when touching generator code and diff the
+  manifests.
+- Downstream conversions must be equally deterministic in CONTENT (same
+  rows, same values); byte-identity of database files is not required
+  (SQLite/PostgreSQL internals may differ), which is why their gate is
+  row counts + query-result checksums instead
+  (see [test-pipeline-validation.md](test-pipeline-validation.md)).
+
+## 6. Anti-patterns
+
+- **Hardcoding `20260711`** in a script instead of reading the config.
+- **An unseeded or globally-seeded random call** anywhere in generation.
+- **`datetime.now()` in corpus values** - simulated schedule only.
+- **Ordering that depends on the filesystem or hash salting** - sort
+  explicitly.
+- **"Harmless" float repr changes** - a formatting change is a corpus
+  change; it must be deliberate and re-baselines every checksum.
+- **Regenerating the corpus to "refresh" it** - identical input config
+  must yield identical output; if bytes changed without a config change,
+  that is a determinism bug to fix, not a new baseline to accept.

docs/rules/data-naming-units.md

@@ -0,0 +1,122 @@
+# Data / Naming and units - IDs, timestamps, SI-explicit column names
+
+**Purpose + scope.** The naming grammar for entities, timestamps, and
+measurement columns across ALL representations of the corpus: CSV, JSON-LD,
+SQLite, PostgreSQL, RDF. Binding for generation, conversion, queries, and
+reporting. This file also fixes two points the specs leave ambiguous
+(sections 3 and 4).
+
+---
+
+## 1. Entity identifiers
+
+The hierarchy encodes containment left-to-right; each level appends one
+segment:
+
+| Entity | Code grammar | Example |
+|---|---|---|
+| batch | `B<nnn>` | `B721` |
+| wafer | `<batch>-W<n>` | `B721-W1` |
+| coupon | `<wafer>-C<nn>` (01..49, the 7x7 grid) | `B721-W1-C07` |
+| track | `<coupon>-T<n>` (1..3) | `B721-W1-C07-T2` |
+| run | `R<n>` (1..4) | `R3` |
+
+- **CSV** stores the bare code (`B721-W1-C07`) in `<entity>_code`-style
+  columns.
+- **JSON-LD / RDF** use the prefixed CURIE form as the node identity:
+  `batch:B721`, `wafer:B721-W1`, `coupon:B721-W1-C07`,
+  `track:B721-W1-C07-T2`, `run:R1` (spec 00_PLAN global conventions).
+- **SQL** stores the bare code as the `<singular>_code` TEXT UNIQUE column
+  next to the integer surrogate key
+  (see [db-sql-schema.md](db-sql-schema.md)).
+- A code is never parsed to recover its parent where a proper reference
+  exists; the parent relation is explicit in every format (directory
+  nesting, `partOf`, FK).
+
+## 2. Timestamps
+
+- ISO-8601, UTC, `Z` suffix, second precision:
+  `2026-03-14T09:30:00Z`.
+- All corpus timestamps are simulated
+  (see [data-determinism.md](data-determinism.md) section 3).
+- In SQL: `TIMESTAMPTZ` in PostgreSQL; TEXT in ISO-8601 form in SQLite
+  (sortable lexicographically by construction).
+
+## 3. Measurement columns - unit suffix is part of the name
+
+Every numeric measurement column ends in an explicit SI unit token. The
+canonical tokens (as they appear in CSV headers):
+
+| Token | Unit | Example columns |
+|---|---|---|
+| `_um` | micrometre | `thickness_um`, `position_um` |
+| `_nm` | nanometre | `ra_nm` |
+| `_GPa` | gigapascal | `hardness_GPa`, `reduced_modulus_GPa` |
+| `_mN` | millinewton | `max_load_mN`, `friction_force_mN` |
+| `_wtpct` | weight percent | `au_wtpct` |
+| `_um3` | cubic micrometre | `wear_volume_um3` |
+| `_m` | metre | `sliding_distance_m` |
+| `_mm_s` | millimetre per second | `speed_mm_s` |
+| `_V` | volt | `discharge_voltage_V` |
+| `_W` | watt | `power_W` |
+| `_C` | degree Celsius | `temperature_C` |
+| `_deg` | degree (angle) | `angle_deg`, `pt_gun_tilt_deg` |
+| `_eV` | electronvolt | `energy_eV` |
+| `_pct` | percent (non-composition) | `rh_pct`, `cpu_util_pct` |
+| `_s` / `_mb` | seconds / megabytes (benchmark outputs) | `wall_s`, `peak_rss_mb` |
+
+- Unit tokens are plain ASCII - never `µ`, `°`, or Unicode superscripts
+  (see [code-data-formatting.md](code-data-formatting.md)).
+- Dimensionless quantities carry no token (`cof`, `cycle`, `k_archard`).
+- New measurement columns pick from this table or add a row to it in the
+  same commit.
+
+## 4. Canonical case mapping CSV -> SQL (ambiguity resolved)
+
+The specs write CSV headers with SI capitalization (`hardness_GPa`, spec
+03) but SQL DDL in lowercase (`hardness_gpa`, spec 05). Binding
+resolution:
+
+- **CSV headers keep SI capitalization** (`_GPa`, `_mN`, `_V`, `_W`,
+  `_C`) - they mirror instrument output conventions.
+- **SQL / JSON-LD / RDF property names are the plain lowercase of the CSV
+  name** (`hardness_gpa`, `max_load_mn`). One mechanical rule, no
+  per-column table to maintain.
+- Converters map with `csv_name.lower()` - never hand-maintained rename
+  dicts that can drift.
+
+## 5. `au_wtpct_mean` (spec gap closed)
+
+Specs 05/06 index `coupons(au_wtpct_mean)` but spec 03 never names the
+column. Binding definition:
+
+- `au_wtpct_mean` = the arithmetic mean of `au_wtpct` over the coupon's
+  400 XRF map points, computed by task 03.
+- It is written into `coupon_info.csv` by generation (not derived later
+  by converters), so every format inherits the identical value, and Q2
+  ("coupons with mean Au = 10 +/- 0.5 wt%") means the same rows
+  everywhere.
+
+## 6. Benchmark artifact columns
+
+Result files under `./out/bench/` follow the same discipline:
+`results_raw.csv` columns are
+`format, query, run, wall_s, peak_rss_mb, cpu_s, cpu_util_pct, read_mb,
+rows, checksum_ok` (spec 09); `storage_sizes.csv` columns are
+`format, variant, bytes` (see
+[build-pipeline-tasks.md](build-pipeline-tasks.md)). Extending either
+file extends this list in the same commit.
+
+## 7. Anti-patterns
+
+- **A measurement column without a unit token** (`thickness`, `load`) -
+  unreadable and un-greppable across formats.
+- **Unit in a comment instead of the name** - comments do not survive
+  format conversion; names do.
+- **Parsing `B721-W1-C07` to find the wafer** where a real reference
+  exists in the format at hand.
+- **A hand-maintained CSV->SQL rename mapping** - the mapping is
+  `lower()`, nothing else.
+- **Recomputing `au_wtpct_mean` in a converter** - it is generated once
+  into `coupon_info.csv`; converters copy it.
+- **Unicode units in identifiers** (`µm`, `°C`) - ASCII tokens only.

docs/rules/db-sql-schema.md

@@ -0,0 +1,139 @@
+# DB / SQL / Schema - conventions for the SQLite and PostgreSQL benchmark targets
+
+**Purpose.** Naming, key, and loading conventions for the two relational
+storage formats under evaluation: the SQLite file (`./out/sqlite/tribo.db`,
+spec [05](../specs/05_convert_sqlite.md)) and the PostgreSQL 16 database
+`tribo` (spec [06](../specs/06_convert_postgresql.md)).
+**Binding scope.** Every table, index, and load routine in either engine.
+The two engines carry the SAME logical schema; they differ only in types,
+storage options, and load mechanics. The full table list and DDL signatures
+live in the specs - this file codifies the conventions the DDL must follow,
+and it wins where a spec detail is ambiguous.
+
+---
+
+## 1. Two kinds of tables
+
+- **Dimension tables** (`batches`, `wafers`, `coupons`, `runs`,
+  `instruments`, `tracks`): small, joined constantly.
+  - PK: integer surrogate, named `<singular>_id` (`coupon_id`, `track_id`).
+    Never a bare `id`, never TEXT.
+  - The natural key from the CSV corpus (`B721`, `B721-W1-C07`,
+    `B721-W1-C07-T2` - see [data-naming-units.md](data-naming-units.md))
+    is stored as a TEXT column named `<singular>_code`, declared `UNIQUE`.
+    It is for lookups and human-readable output, never the join key.
+- **Bulk tables** (`friction_cycles`, `friction_loop_points`, `xrf_points`,
+  `profilometry_points`, `nanoindentation`, `afm`, `wear`): millions of
+  rows, scanned and filtered.
+  - Reference dimensions by integer FK columns only. **TEXT foreign keys in
+    bulk tables are forbidden** - they bloat the file and slow every scan.
+  - PK is the natural composite where one exists: `(track_id, cycle)` for
+    cycles, `(track_id, cycle, pt)` for loop points; single-column
+    `track_id` / `coupon_id` where the grain is one row per entity. In
+    SQLite, composite-PK bulk tables are `WITHOUT ROWID`.
+  - No surrogate `bigserial` on bulk tables - it adds bytes and nothing
+    else.
+
+## 2. Table and column naming
+
+- Tables: lowercase, snake_case, plural (`friction_cycles`, not
+  `FrictionCycle`).
+- Columns: lowercase snake_case; measurement columns carry the SI unit
+  suffix in **lowercase** (`hardness_gpa`, `max_load_mn`,
+  `wear_volume_um3`). The CSV corpus keeps SI capitalization
+  (`hardness_GPa`); the SQL mapping is a plain lowercase of the CSV name -
+  see the canonical mapping in
+  [data-naming-units.md](data-naming-units.md).
+- Denormalization is deliberate and limited: `environment` and `load_mn`
+  are copied onto `tracks` (from `runs` / test conditions) because
+  Q4 and Q7 filter on them constantly. Do not denormalize further without
+  updating this rule.
+
+## 3. Indexes - only what Q1-Q7 use
+
+- Create indexes AFTER the bulk load, never before.
+- The index set is driven by the benchmark queries: `coupons(au_wtpct_mean)`,
+  `tracks(run_id)`, `tracks(coupon_id)`, `runs(environment)`, composite
+  `tracks(environment, load_mn)`; PostgreSQL adds BRIN on
+  `friction_cycles(cycle)` per partition.
+- Naming: `ux_<table>_<cols>` for unique, `ix_<table>_<cols>` for
+  non-unique.
+- Do not over-index: an index no query uses is disk footprint that skews
+  the storage comparison AND load-time cost. Every index must name the
+  query it serves (a one-line comment in the DDL).
+
+## 4. Loading
+
+- **SQLite:** during load `PRAGMA journal_mode=OFF, synchronous=OFF,
+  cache_size=-262144`; insert via `executemany` in transactions of 50k
+  rows; after load `PRAGMA journal_mode=WAL`, `ANALYZE`, `VACUUM`.
+- **PostgreSQL:** load via `COPY FROM STDIN` (psycopg3 `copy`), never
+  row-by-row INSERT; `friction_cycles` and `friction_loop_points` are
+  declaratively RANGE-partitioned by `track_id` (12 partitions of 120
+  tracks); `VACUUM ANALYZE` after load.
+- Loads are all-or-nothing per table and validated against `MANIFEST.csv`
+  row counts - see [code-error-handling.md](code-error-handling.md)
+  section 3 and [test-pipeline-validation.md](test-pipeline-validation.md).
+- **Foreign keys are ENFORCED in both engines on every table, including
+  bulk tables** (owner decision 2026-07-11, after measurement on the real
+  corpus: +0.5 s load on 4.3M rows in SQLite, zero disk and read cost;
+  full study with ER diagrams and measurements:
+  [docs/research/Tribology_FK_Architecture.html](../research/Tribology_FK_Architecture.html)).
+  Every SQLite connection sets `PRAGMA foreign_keys = ON` (SQLite silently
+  ignores FK clauses without it). Row-count validation against
+  `MANIFEST.csv` remains a second, independent integrity gate.
+
+## 5. `track_summary` - precomputed, and off-limits to Q5
+
+Both engines carry a `track_summary` (plain table computed during load in
+SQLite; materialized view in PostgreSQL) with per-track derived values
+(`cof_ss_mean`, `run_in_cycles`, ...). It exists for Q2/Q3/Q4/Q6. **Q5 must
+never read it** - Q5 is the forced full-scan benchmark over raw
+`friction_cycles` in every format (spec
+[08](../specs/08_benchmark_queries.md)).
+
+Binding computation (identical semantics in both engines; float results
+agree within the 1e-9 checksum tolerance):
+
+1. Steady-state proxy window = cycles 501-1000 (the second half);
+   `m` = avg(cof), `s` = stddev_pop(cof) over that window.
+2. `run_in_cycles` = the first cycle `c` with `|cof(c) - m| < 2*s`;
+   fallback 500 if no cycle qualifies.
+3. `cof_ss_mean` / `cof_ss_std` = avg / stddev_pop of cof over cycles
+   `> run_in_cycles`.
+
+SQLite computes this in Python during load
+([common/track_summary.py](../../common/track_summary.py)); PostgreSQL as
+a materialized view implementing the same three steps.
+
+## 6. Engine parity
+
+- Same logical schema, same table names, same column names, same row
+  counts in both engines.
+- Q1-Q7 executed against SQLite and PostgreSQL MUST return identical
+  result sets (checksum parity with 1e-9 float tolerance - the canonical
+  expected checksums are computed from PostgreSQL and cross-validated
+  against SQLite; see
+  [test-pipeline-validation.md](test-pipeline-validation.md)).
+- A schema change in one engine is a schema change in both, in the same
+  commit.
+
+## 7. Anti-patterns
+
+- **Bare `id` PK, TEXT PK, or a GUID column** - this is a load-once
+  analytical schema; surrogate integer `<singular>_id` on dimensions,
+  natural composites on bulk tables.
+- **TEXT FK in a bulk table** (`friction_cycles.track_code`) - integer FKs
+  only.
+- **A `bigserial` surrogate on `friction_cycles` / `friction_loop_points`**
+  - dead weight at 1.4M / 2.9M rows.
+- **Creating indexes before the bulk load** - doubles load time for
+  nothing.
+- **An index without a named query** - footprint that skews the format
+  comparison.
+- **Rewriting Q5 to use `track_summary`** - breaks the benchmark's
+  full-scan contract.
+- **`updated_at` / audit columns / triggers** - this is a read-only
+  benchmark target rebuilt from CSV; it has no mutation lifecycle.
+- **Divergent schemas between the engines** - parity is what makes the
+  benchmark comparison valid.

docs/rules/meta-concise-answers.md

@@ -0,0 +1,59 @@
+# Meta / Concise answers
+
+**Binding on every response.** Match length to the question. Stop after the answer.
+
+## Rule
+
+Do not append: background, rationale, anticipated follow-ups, recap, summary,
+TL;DR, "why this matters", offers to do more. If the user said "concise",
+"short", "one line", or "just" - obey literally.
+
+## Length budget
+
+| Question shape | Max response |
+|---|---|
+| "What is X?" / "Where is Y?" / yes-no | 1 line |
+| "How do I do X?" | 1-3 lines, commands only |
+| "How do I test X?" | 3-5 lines, commands + check |
+| "Show me X" (file, diagram, output) | The thing itself, no preamble |
+| "Compare A and B" | 1 short table or 2-3 bullets/side |
+| Open-ended architecture / discussion | Full prose acceptable |
+
+Diagrams / lists / tables: **8-line ceiling** unless user asks for more.
+No ASCII box-drawing unless asked. When the user says "I don't care about X" -
+omit X, do not list it "for completeness".
+
+## Examples
+
+**Bad** - user asks "what does this option do?", assistant returns 6-option
+table with pros/cons. **Good** - one sentence about the one option.
+
+**Bad** - user asks "show ER diagram, no fields", assistant returns 80-line
+ASCII with cardinality table and "what changes" section. **Good** - 4-5
+lines of `entity --< child` arrows for the asked entities only.
+
+**Bad** - yes/no question answered with a paragraph of nuance before the
+yes/no. **Good** - yes/no first, nuance only if asked.
+
+## Exceptions (these ARE the answer, no length cap)
+
+File contents requested, code requested, test/log output, final summary
+after multi-step tool work, plans the user asked to be detailed.
+
+## Self-check before sending
+
+- Did the user ask a yes/no? -> answer yes/no in the first line, nuance
+  only after, only if useful.
+- Did the user ask for one thing? -> deliver that one thing only.
+- About to write more than ~10 lines for a question that needs ~5? -> cut
+  everything past line 10 unless it is code, output, or a file path.
+- Did the user say "concise" / "short" / "one line" / "just"? -> obey
+  literally: one sentence or one command/path.
+
+## Recovery
+
+If the user complains the output is too long:
+
+1. Acknowledge in one line.
+2. Re-answer the original question in one line.
+3. Capture it as an anti-pattern in the Examples section above.

docs/rules/meta-concise-questions.md

@@ -0,0 +1,33 @@
+# Ask extremely short, concise questions
+
+Binding on EVERY question put to the user.
+
+## Rules
+
+- **One decision per question.** Do not bundle unrelated questions.
+- **Minimal context.** State only what is needed to answer: the situation in one
+  short clause, the concrete options, and what the answer changes. Nothing else.
+- **No preamble.** Do NOT precede a question with analysis, restated reasoning,
+  background, evidence, or a summary of what you just did. The user has that
+  context; if a fact is truly needed to decide, compress it to one clause.
+- **Prefer the smallest form.** A yes/no, or a 2-3 option pick. Put the question
+  itself first or make it unmissable.
+- **Self-contained but short.** The options must be understandable without opening
+  another document, but spell them out in a few words, not paragraphs.
+- **No useless information.** Every word must help the user decide. Delete anything
+  that does not.
+
+## Anti-patterns
+
+- A multi-paragraph lead-in before the actual question.
+- Re-explaining the problem the user already stated.
+- Listing options with a paragraph of trade-offs each.
+- Asking two or more separate decisions in one run of prose.
+
+## Shape to aim for
+
+```
+<one-clause situation>. <Question>?
+1. <option> - <few words>
+2. <option> - <few words>
+```

docs/rules/meta-cursor-rules.md

@@ -0,0 +1,66 @@
+# Meta / Cursor rule artifact - commit-message generation
+
+**Binding** for keeping Cursor's commit-message generation aligned with this
+project's commit convention.
+
+The development environment is the Cursor IDE with the Claude Code extension.
+Claude Code reads the root `CLAUDE.md` and `docs/rules/` directly and needs no
+`.cursorrules`. Cursor's OWN built-in AI (Chat/Composer and the SCM "sparkle"
+commit-message generator) does not read those files - it reads `.cursorrules`.
+Since the user creates all commits themselves (see
+[`meta-git-commits.md`](meta-git-commits.md)) and may use the one-click SCM
+generator to draft the message, `.cursorrules` must exist and must encode the
+commit convention.
+
+This project ships a SINGLE Cursor rules file: the root
+[`.cursorrules`](../../.cursorrules). There is no `.cursor/rules/*.mdc`
+directory here, and none should be added. All Cursor rule content lives in
+`.cursorrules`.
+
+---
+
+## 1. The requirement
+
+`.cursorrules` MUST encode this project's commit-message convention so that
+Cursor's generator emits conforming messages. The authoritative convention
+lives in [`meta-git-commits.md`](meta-git-commits.md); `.cursorrules` mirrors
+it as plain-text instructions to the generator.
+
+The failure mode this rule prevents: a missing or stale `.cursorrules`, so the
+one-click SCM generator falls back to Cursor defaults and emits unprefixed junk
+(`Update files`, `Improve project`).
+
+## 2. What `.cursorrules` must encode
+
+1. **First-line format** - `type(scope): description`, lowercase imperative
+   summary in English, under 72 chars, no trailing period.
+2. **Allowed types (closed list)** - `feat fix perf refactor security docs ci
+   chore test`.
+3. **Allowed scopes (closed list)** - `specs rules docs config datagen convert
+   bench report tools all`, with the area mapping from
+   [`meta-git-commits.md`](meta-git-commits.md).
+4. **Coverage rule** - the subject is only a title; multi-area diffs need a
+   blank line and a 2-6 bullet body covering every meaningful changed area; use
+   `all` when no narrower scope honestly fits; split unrelated work.
+5. **Anti-examples** - forbid vague/unprefixed lines (`Update files`, `Improve
+   project`, `clarify rules`).
+6. **Allowed special forms** - `Merge ...` commits are allowed as-is;
+   `release(scope): vX.Y.Z` is allowed for releases.
+7. **Staged-diff grounding** - base the message ONLY on `git diff --cached`;
+   mention only staged paths; if the index is empty, output `No staged
+   changes`; never infer from unstaged diffs, untracked files, prior
+   commits, or chat context.
+
+## 3. Keeping the mirror in sync
+
+No commit-msg hook is installed in this repository; the convention is enforced
+by review. `.cursorrules` and [`meta-git-commits.md`](meta-git-commits.md)
+must agree: same type list, same scope list, same first-line format. If the
+convention changes, update both files in the same change set.
+
+## 4. Scope boundary
+
+This rule governs the commit-message *convention materialization* - not who may
+create commits. Commit authorship is governed by
+[`meta-git-commits.md`](meta-git-commits.md): rules files describe the
+convention; only the user creates actual commits.

docs/rules/meta-git-commits.md

@@ -0,0 +1,99 @@
+# Meta / Git commits - never on behalf of the user
+
+**Binding** for every contributor (human or AI agent) touching this
+repository. Applies to **all** git history-modifying operations.
+
+---
+
+## 1. The rule
+
+**Never create a git commit on behalf of the user. Ever.** The user is the
+only entity allowed to materialize a commit in this repo.
+
+This covers, without exception:
+
+- `git commit` (with or without `-a`, `-m`, `--amend`)
+- `git merge` - never finalize one
+- `git rebase` (interactive or otherwise) - never start one
+- `git revert` - never run; the user decides whether a change is undone
+- `git cherry-pick`
+- `git stash`
+- `git tag` (annotated or lightweight)
+- `git push` of any kind - pushing is the user's call
+- `git reset` / `git checkout` that changes the working tree or refs
+- Any wrapper script or `gh` API call that would create a commit / tag /
+  merge on the user's behalf
+
+For an AI agent, git is **read-only**: reading history (`git log`,
+`git diff`, `git status`, `git show`) and inspecting branches is fine;
+anything that writes to the index, refs, or `objects/` - including
+`git add` - is left to the user. (This matches the machine-level
+enforcement: git write commands by the agent are denied by a hook.)
+
+## 2. Why
+
+- Commit authorship and message wording are decisions the user owns - they are
+  durable, public, and signed against the user's name and email.
+- "Tiny" commits that look harmless ("fix typo", "delete unused file") still
+  rewrite history and cannot be quietly reversed once pushed.
+- Generated data and measured results in this project take hours to rebuild;
+  what enters history and when is a conscious decision, not a side effect.
+
+## 3. What to do instead
+
+- Make the working-tree changes.
+- Report what changed and offer a draft commit message **as plain text** in the
+  conversation, together with the exact `git add` / `git commit` commands for
+  the user to run themselves.
+
+## 4. Commit-message format
+
+When you draft a message, follow this project's convention:
+
+- **First line** - `type(scope): description`, lowercase imperative summary in
+  English, under 72 chars, no trailing period.
+- **Types (closed list)** - `feat fix perf refactor security docs ci chore
+  test`.
+- **Scopes (closed list, owner-approved 2026-07-11)** - `specs rules docs
+  config datagen convert bench report tools all`. Pick by the area of the
+  changed files:
+  - `specs` = `docs/specs/**`
+  - `rules` = `docs/rules/**`
+  - `docs` = other prose docs (`README.md`, root `CLAUDE.md`,
+    `docs/examples/**`)
+  - `config` = configuration files (`.gitignore`, `.cursorrules`, editable
+    YAML defaults such as hardware prices)
+  - `datagen` = lab-configuration and data-generation code (tasks 01-03:
+    `lab_config` generation, process diagrams, `generate_data.py`)
+  - `convert` = format converters (tasks 04-07: JSON-LD, SQLite,
+    PostgreSQL, RDF)
+  - `bench` = benchmark queries, runner, and extrapolation (tasks 08-10)
+  - `report` = reporting code and templates (task 11)
+  - `tools` = helper scripts under `tools/`
+  - `all` = multiple areas or in doubt
+- **Coverage** - multi-area diffs get a blank line and a 2-6 bullet body
+  covering every meaningful changed area; split unrelated work.
+- **Special forms** - `Merge ...` and `release(scope): vX.Y.Z` are allowed
+  as-is.
+
+No `commit-msg` hook is installed in this repository yet; the convention is
+enforced by review. If a hook is added later, it must implement exactly this
+format and this file must be updated in the same change. The same convention
+is mirrored into `.cursorrules` for Cursor's built-in commit-message
+generator - see [meta-cursor-rules.md](meta-cursor-rules.md).
+
+## 5. Anti-patterns
+
+- **"I'll just commit it so we have a checkpoint."** No. Working-tree state is
+  the checkpoint; the user decides when it becomes history.
+- **Amending the user's last commit to fix something you noticed.** No. Surface
+  the issue, let the user choose between amend and new commit.
+- **Suggesting `git add -A` in the draft commands.** List explicitly named
+  files, so the user sees exactly what would land.
+- **Vague subjects** (`Update files`, `Improve project`, `clarify rules`).
+  Say what changed and where.
+
+## 6. Exception (closed list)
+
+There are no exceptions. If a workflow seems to require a commit, pause,
+explain the dependency, and let the user run the git command.

docs/rules/meta-language-style.md

@@ -0,0 +1,35 @@
+# Meta / Communication language and register
+
+**Binding on every response and on all prose directed at the user**
+(explanations, status updates, questions, summaries, commit messages, PR text).
+
+## Rule
+
+1. **Language: Russian.** All communication with the user is in standard
+   literary Russian.
+2. **Register: neutral and professional. No jargon, no slang, no profanity.**
+   - No profanity or vulgar words - in any language, under any circumstances.
+   - No slang or jargon: street slang, gamer/internet slang, or untranslated
+     English buzzwords used as slang.
+   - Prefer the established Russian word over a transliterated English one
+     whenever a normal Russian equivalent exists.
+   - Plain, grammatical, complete sentences.
+
+## Allowed terminology (not slang)
+
+Field-standard technical terms are fine - they are terminology, not jargon:
+e.g. «коммит», «билд», «дебаг», «релиз», «мерж». When unsure whether a word
+is accepted terminology or slang, choose the neutral Russian equivalent.
+
+## Scope
+
+Does **not** apply to source code, identifiers, file paths, code comments,
+shell/tool output, or verbatim quotes of errors and logs - those keep the
+language of the codebase.
+
+## Canonical source
+
+This file is the project-level binding rule. It restates the project owner's
+standing preference (communication in Russian; code, docs, and artifacts in
+English). If a global user-level rule ever diverges from this file, the
+stricter wording wins.

docs/rules/meta-no-assumptions.md

@@ -0,0 +1,51 @@
+# Meta / No assumptions
+
+**Binding** on every interaction. Strictest when the cost of a wrong
+guess is high: deleting or regenerating `./out/` artifacts, dropping or
+reloading the `tribo` PostgreSQL database, changing the random seed or
+`lab_config.yaml`, editing specs in ways that invalidate generated data,
+git history, anything with data-loss consequences.
+
+## Rule
+
+Before producing a plan or change:
+
+1. Re-read the user's most recent message. Watch for enumerations
+   ("both", "and also", "in addition to"), explicit purpose statements
+   ("X for A, Y for B"), and distinct entities that sound similar.
+2. Don't collapse two distinct entities into one. "X for A and Y for B"
+   means **two** setups, not one.
+3. If anything is ambiguous after careful re-reading - **ask**. Don't
+   pick the most likely interpretation and proceed.
+4. Echo back the parsed interpretation before producing the plan, so a
+   misread is cheap to correct.
+
+## When the rule is most strict
+
+Strictness scales with the cost of a wrong guess:
+
+| Domain | Why a wrong assumption is expensive |
+|---|---|
+| **Deleting / regenerating `./out/`** | The artifact tree holds hours of generation, conversion, and benchmark work. Regenerating CSV invalidates every downstream format, benchmark result, and report. Confirm which subtree is affected before deleting or overwriting. |
+| **The `tribo` PostgreSQL database** | Dropping or reloading it discards a long `COPY` load and invalidates benchmark results derived from it. Confirm before any `DROP` / `TRUNCATE` / reload. |
+| **The random seed / `lab_config.yaml`** | Seed 20260711 and the lab configuration are the root of the whole reproducibility chain. Changing either silently invalidates all generated data, all conversions, all checksums, and all measured results. Never change them as a side effect of another task. |
+| **Editing specs (`docs/specs/`)** | Specs are the task contract. A "small clarification" that changes volumes, schemas, or file layouts desynchronizes already-generated artifacts. Confirm intent when an edit is more than wording. |
+| **Benchmark results (`./out/bench/`)** | Measured results cannot be re-derived without re-running the full matrix (hours). Never overwrite `results_raw.csv` / `results_median.csv` without confirmation. |
+| **git history** | A wrong rebase/reset/force-push rewrites shared history; recovery is manual and error-prone. (The AI agent never runs git write operations at all - see [meta-git-commits.md](meta-git-commits.md).) |
+
+For all of the above: **the cost of one extra clarifying question is
+zero; the cost of a wrong assumption is hours-to-days** of recovery.
+Always ask.
+
+## Forms a clarifying question takes
+
+- "Confirm: you want X for A AND Y for B (two setups), not just one?"
+- "Two readings: (a) ..., (b) ... - which?"
+- "What does '<word>' refer to here - (a) ... or (b) ...?"
+
+## When the assumption was already wrong
+
+1. Acknowledge the misread explicitly. Quote the user's words back.
+2. Discard the wrong artifact entirely; don't repair piecemeal unless
+   the user asks.
+3. Restart with the corrected interpretation, echo it first.

docs/rules/meta-rules-naming.md

@@ -0,0 +1,119 @@
+# Meta / Rule file naming
+
+**Binding** for files inside [`docs/rules/`](docs/rules/). Adding, renaming, or
+splitting a rule file follows this document.
+
+## Layout
+
+- `docs/rules/` - binding rules (this folder): reproducible conventions on HOW
+  work is done. Reference content that supports a rule also lives here,
+  named `<rule>-reference.md`.
+- `docs/specs/` - the pipeline task specifications (WHAT to build: tasks
+  01-11, their inputs, outputs, and dependencies). Binding for scope and
+  deliverables, but conventions that span tasks are codified in `docs/rules/`.
+- `docs/examples/` - imported reference materials (master plan, ontology).
+  Descriptive, not binding.
+
+Files live flat at the top level of `docs/rules/`. No subfolders.
+
+## Naming
+
+```
+<domain>-<subdomain>-<topic>[-<aspect>].md
+```
+
+- Lowercase, dashes between segments. No underscores, no CamelCase.
+- Max 4 segments. Most general first.
+- Reading the filename should answer: category -> subcategory -> topic.
+
+## Approved domain prefixes
+
+| Prefix | Covers |
+|---|---|
+| `bench-` | Benchmark execution, resource measurement, extrapolation, reporting methodology |
+| `build-` | Pipeline task orchestration: the `./out/` tree, `.done` markers, task order |
+| `code-` | General code style across languages (Python, SQL, PowerShell): comments, formatting, error handling, configuration |
+| `data-` | The simulated dataset: determinism, identifiers, units, column naming |
+| `db-` | SQLite / PostgreSQL benchmark target schemas, loading, indexing |
+| `meta-` | Rules about the rules system and working conventions |
+| `obs-` | Observability and logging |
+| `test-` | Validation gates, checksums, pytest conventions |
+
+Don't invent a new prefix without adding it to this table in the same
+commit.
+
+### Subdomains (second / third segment)
+
+Within a domain, the second segment is a recognised tool, subsystem, or
+concern; the third is the concrete topic. Established families:
+
+- `db-sql-*` - relational benchmark targets (SQLite, PostgreSQL): schema,
+  loading, indexing.
+- `data-*` - dataset conventions (`data-determinism`, `data-naming-units`).
+- `bench-*` - measurement methodology (`bench-methodology`).
+- `build-pipeline-*` - task orchestration and artifact layout.
+- `code-*` - general code style (comments, formatting, error handling,
+  config, Python style).
+- `meta-*` - rules about the rules system and working conventions.
+- `obs-*` - observability and logging (`obs-python-logging`).
+- `test-*` - validation and testing (`test-pipeline-validation`).
+
+## Examples
+
+| Good | Reads as |
+|---|---|
+| `data-determinism.md` | Data -> deterministic generation |
+| `db-sql-schema.md` | DB -> SQL -> schema conventions |
+| `bench-methodology.md` | Benchmarks -> measurement methodology |
+| `code-python-style.md` | Code -> Python -> style |
+| `obs-python-logging.md` | Observability -> Python logging |
+| `meta-rules-naming.md` | Meta -> rules system -> naming |
+
+| Bad | Why |
+|---|---|
+| `BenchMethodology.md` | CamelCase forbidden |
+| `data_determinism.md` | Underscores forbidden |
+| `sqlite-loading-conventions.md` | Use `db-` prefix, drop redundant `conventions` |
+| `bench-execution-measurement-rules-doc.md` | 5+ segments + redundant `rules` suffix |
+
+## Adding a rule
+
+1. Pick the domain prefix; add new ones to the table above first.
+2. Check `ls docs/rules/` - extend a sibling rule before creating a new one.
+3. Create the file with binding content, examples, and exceptions.
+4. Add a row for the new file to the rule-trigger table in the root
+   [`CLAUDE.md`](../../CLAUDE.md) in the same commit.
+5. If replacing an older doc, `git mv` to preserve history and `Grep` the
+   repo for the old path.
+
+## Rules vs specs vs memory files
+
+Three homes for a statement; pick by its kind:
+
+| Source | Stores |
+|---|---|
+| `docs/rules/<name>.md` | A reproducible, mandatory behaviour - the same answer for every contributor on every task within the topic. |
+| `docs/specs/NN_<task>.md` | The definition of one pipeline task: scope, inputs, outputs, dependencies. |
+| `memory/<name>.md` | A fact about *this user*, *this project*, or *the current moment* - preferences, ongoing initiatives, locked decisions, infra specifics. |
+
+- A convention that must hold across several tasks -> `docs/rules/`.
+- A deliverable of one task -> its spec in `docs/specs/`.
+- A fact about the user or the present state of things -> `memory/`.
+
+A rule file may reference a spec or a memory note for context, but the
+binding convention lives in the rule file. When a rule and a spec disagree
+on a convention (naming, formats, thresholds), the rule wins and the spec
+is updated to match.
+
+## Anti-patterns
+
+- **Duplicating rule content across rule files** - pick one canonical
+  location and cross-link from siblings instead of copying.
+- **Restating a whole spec inside a rule** - the rule codifies the
+  reusable convention and links the spec for task detail.
+- **Adding rules to the body of CLAUDE.md** instead of creating a rule
+  file - CLAUDE.md carries only the trigger table, the project preamble,
+  and the canonical-source-of-truth note.
+- **Naming a file after the implementation** instead of the domain
+  (`oxigraph-loading.md` vs `db-rdf-loading.md`) - the implementation may
+  be replaced; the topic endures.

docs/rules/obs-python-logging.md

@@ -0,0 +1,113 @@
+# Observability / Python logging - standard library only, namespaced, plain text
+
+**Purpose + scope.** Every Python module in this repository (data generation,
+converters, benchmark runner, extrapolation, reporting, helper tools) MUST
+use the standard-library `logging` module following these conventions. No
+external logging dependency.
+
+---
+
+## 1. The rule
+
+1. **Standard library `logging` only.** No `loguru`, no `structlog`, no
+   `python-json-logger`. Stdlib is sufficient and keeps the dependency list
+   closed (see [code-python-style.md](code-python-style.md)).
+2. **Namespaced loggers.** Every module obtains its logger via
+   `logging.getLogger(__name__)`. Never the root logger directly.
+3. **`print()` is for the user-facing CLI protocol only.** The specs
+   explicitly require certain stdout output: progress summaries, the real
+   generated tree size (task 03), per-cell benchmark progress, final
+   counts. That protocol output goes through `print()`. Everything
+   diagnostic - internal flow, warnings, errors, per-row detail - goes
+   through a logger.
+4. **Log level from a CLI flag.** Every pipeline script accepts
+   `--log-level {DEBUG,INFO,WARNING,ERROR}`, default `INFO`. There is no
+   settings table and no `.env` in this project (see
+   [code-config-yaml.md](code-config-yaml.md)).
+5. **Plain-text format on stderr.** These are one-shot CLI scripts, not
+   services; JSON logs are overkill. One line per record via
+   `logging.basicConfig`. Keeping the logger on stderr leaves stdout clean
+   for the protocol output of item 3.
+6. **Never log secrets.** The only credential material in this project is
+   the PostgreSQL DSN - never log it (a DSN may embed a password). Log the
+   database name or host, never the full connection string.
+
+## 2. API by severity
+
+| Method | Use for |
+|---|---|
+| `logger.critical(...)` | The process cannot continue - before `sys.exit(1)` or an unrecoverable state. |
+| `logger.error(...)` | Failed operation, exception caught and handled, broken invariant, a benchmark cell recorded as TIMEOUT/ERROR. Always include `exc_info=True` when an exception is in scope. |
+| `logger.warning(...)` | Recoverable problem, retry triggered, suspicious state (e.g. a value near a validation bound), an optional step skipped (e.g. cold-cache pass unavailable). |
+| `logger.info(...)` | Normal flow milestones - task start/finish, per-batch progress, row counts written, validation gates passed. |
+| `logger.debug(...)` | Verbose tracing - per-coupon detail, generated value samples, SQL statements, per-run raw measurements. Hidden at default `INFO`. |
+
+**Default level: `INFO`.** `DEBUG` is opt-in via `--log-level DEBUG` for a
+diagnosis session.
+
+## 3. Logger acquisition and setup
+
+```python
+# Top of every module that logs:
+import logging
+logger = logging.getLogger(__name__)
+
+# Once, in each script's main():
+logging.basicConfig(
+	level=args.log_level,
+	stream=sys.stderr,
+	format="%(asctime)s %(levelname)-7s %(name)s: %(message)s",
+)
+```
+
+Inside functions:
+
+```python
+def load_table(conn, name: str, rows) -> int:
+	logger.info("loading %s", name)
+	try:
+		...
+	except psycopg.Error:
+		logger.error("loading %s failed", name, exc_info=True)
+		raise
+```
+
+`__name__` gives the module path as the namespace tree used for filtering.
+Never hardcode a logger name string. Clamp noisy third-party libraries
+(`psycopg`, `matplotlib`) to `WARNING` in the same setup block if they
+pollute the output.
+
+## 4. Progress reporting in long loops
+
+Generation and loading loops run over millions of rows. Report progress at
+INFO on a coarse cadence (per batch / per wafer / per table, or every N
+seconds), never per row. Per-row output belongs at DEBUG, and even there
+prefer sampled output ("first row of each file") over full dumps.
+
+## 5. Anti-patterns
+
+- **`print()` for diagnostics** - forbidden; `print()` is only the
+  user-facing CLI protocol (progress summaries, final counts, sizes).
+- **Diagnostics on stdout** - stdout is the protocol channel; the logger
+  writes to stderr.
+- **Root logger calls** (`logging.info(...)`) - forbidden; always
+  `logger = logging.getLogger(__name__)`.
+- **Third-party logging frameworks** (`loguru`, `structlog`) - forbidden.
+- **Hardcoded logger names** (`logging.getLogger("datagen")`) - forbidden;
+  pass `__name__`.
+- **Logging the PostgreSQL DSN / connection string** - forbidden; log
+  host/database name only.
+- **Log level from an environment variable or a config file** - the level
+  is a per-run diagnostic choice; it comes from `--log-level`.
+- **`logger.error(...)` without re-raise** when the function semantically
+  cannot proceed - see [code-error-handling.md](code-error-handling.md).
+- **Per-row INFO logging in bulk loops** - drowns the signal and slows the
+  run; use coarse milestones.
+
+## 6. Related documents
+
+- [code-error-handling.md](code-error-handling.md) - `exc_info=True`
+  requirement and the swallow ladder.
+- [code-config-yaml.md](code-config-yaml.md) - configuration scope (no
+  `.env`, no settings table).
+- [code-python-style.md](code-python-style.md) - the closed dependency list.

docs/rules/test-pipeline-validation.md

@@ -0,0 +1,125 @@
+# Test / Pipeline validation - markers, manifests, checksums
+
+**Purpose + scope.** The validation contract between pipeline tasks: how a
+task proves it completed correctly and how downstream tasks verify their
+inputs before trusting them. Binding for every task implementation (01-11)
+and for any pytest suite added to this repository.
+
+The pipeline has no human QA step between tasks - these gates are the only
+thing standing between a silent generation/conversion bug and a report full
+of wrong numbers.
+
+---
+
+## 1. Completion markers - written last, verified first
+
+- Every task writes `./out/.done/<NN>.ok` **only after** all its outputs
+  are complete and its own validation passed. The marker contains the
+  task's headline counts (rows, bytes, triples) for downstream checks.
+- A task MUST verify the `.done` markers of its declared dependencies
+  before starting, and abort with a clear message if one is missing.
+- A task that re-runs deletes its own marker first and rewrites it at the
+  end. A marker is never written on failure and never from a `finally`
+  block (see [code-error-handling.md](code-error-handling.md)).
+
+## 2. The CSV corpus manifest - the root of trust
+
+- Task 03 writes `./out/csv/MANIFEST.csv`: one row per generated file with
+  path, row count, byte size, sha256.
+- Every converter (04-07) validates what it read against the manifest: row
+  counts per source file, and sha256 where the file is consumed whole. A
+  mismatch is an abort, not a warning.
+- The manifest is also the reproducibility proof: regenerating with the
+  same `lab_config.yaml` must reproduce identical sha256 values (see
+  [data-determinism.md](data-determinism.md)).
+
+## 3. Canonical expected volumes
+
+The lab configuration fixes the corpus size exactly; validation asserts
+these numbers, not "roughly that many":
+
+| Entity | Count |
+|---|---|
+| batches | 4 |
+| wafers | 12 |
+| coupons | 588 (480 friction + 108 characterization-only) |
+| tracks | 1,440 |
+| friction cycle rows | 1,440,000 |
+| friction loop points | 2,880,000 |
+| XRF points | 235,200 (588 x 400) |
+| generated tree size | 120-200 MB (hard abort bounds) |
+
+If a deliberate `lab_config.yaml` change moves these numbers, this table
+and the validation code change in the same commit.
+
+## 4. Query-result checksums - all formats agree
+
+- Each benchmark query Q1-Q7 has ONE canonical expected result: rows
+  sorted, floats compared with tolerance 1e-9, checksummed to
+  `./out/bench/expected/q<N>.sha256`.
+- The canonical checksums are computed from **PostgreSQL** and
+  cross-validated against **SQLite** before any benchmarking starts; a
+  disagreement between the two engines is a converter or query bug and
+  blocks task 09.
+- Every measured benchmark run records a `checksum_ok` flag; a cell whose
+  result does not match the expected checksum is an invalid measurement
+  regardless of how fast it ran.
+
+## 5. Format smoke tests
+
+Each converter proves basic well-formedness of its output before writing
+its marker:
+
+- **JSON-LD (04):** one sample coupon file passes `json.load` AND an
+  rdflib JSON-LD parse.
+- **SQLite (05) / PostgreSQL (06):** row counts per table match the
+  manifest-derived expectations; `track_summary` row count equals 1,440.
+- **RDF (07):** the loaded store answers a SPARQL `COUNT` matching the
+  serialized triple count recorded in the `.done` marker.
+
+## 6. If a pytest suite is added
+
+The specs define no pytest suite; validation is built into the tasks. If
+unit/integration tests are added later:
+
+- **Skip, don't fail, when PostgreSQL is unreachable.** A checkout without
+  a running PostgreSQL still gets a green (skipped) run;
+  connection failure in a fixture is `pytest.skip(...)`, not an error.
+- **Never mutate real `./out/` artifacts.** Tests that need data build a
+  small corpus into a temp directory (a reduced `lab_config` is fine) or
+  copy the artifact; hours of generated/measured data are not test
+  fixtures to be overwritten.
+- **Ephemeral databases only.** A test touching PostgreSQL creates and
+  drops its own database (e.g. `tribo_test`) next to `tribo`; it never
+  writes to `tribo` itself.
+- **Independent and order-free.** Every test passes run alone; no residue
+  between tests, no execution-order dependencies.
+- **Deterministic assertions.** With the fixed seed, exact counts and
+  exact values are assertable - prefer exact assertions over "greater than
+  zero".
+
+## 7. Anti-patterns
+
+- **Writing a `.done` marker before validation, on failure, or in
+  `finally`** - the marker means "completed and verified".
+- **Skipping manifest validation "because generation just ran"** - the
+  gate exists precisely for the runs where the assumption is wrong.
+- **Warning-and-continuing on a count mismatch** - a converter with a row
+  deficit produces a database that LOOKS complete; abort.
+- **Regenerating expected checksums to make a failing format pass** - the
+  canonical result comes from PostgreSQL cross-validated with SQLite;
+  a mismatch in another format is a bug in that converter or query.
+- **Tests writing into `./out/`** or connecting to the real `tribo`
+  database.
+- **Failing (not skipping) when PostgreSQL is unavailable** in a pytest
+  run.
+
+## 8. Related documents
+
+- [build-pipeline-tasks.md](build-pipeline-tasks.md) - marker layout and
+  task order.
+- [data-determinism.md](data-determinism.md) - why byte-identical
+  regeneration is assertable.
+- [code-error-handling.md](code-error-handling.md) - fail-loud mechanics.
+- [bench-methodology.md](bench-methodology.md) - how `checksum_ok` enters
+  the measurement protocol.

docs/specs/00_PLAN.md

@@ -0,0 +1,55 @@
+# 00_PLAN - Tribology Lab Data Storage Evaluation Pipeline
+
+> Provenance: this plan and the 11 task specifications were generated from
+> [../Initial_Prompt.md](../Initial_Prompt.md). The prompt is a historical
+> artifact, not a contract; the specs and [docs/rules/](../rules/) supersede
+> it wherever they differ.
+
+## Goal
+Build a complete, reproducible pipeline that:
+1. Defines a simulated tribology laboratory (Sandia Pt-Au LDRD context) with real instruments and a realistic test workflow.
+2. Generates ~150 MB of physically plausible measurement data in CSV (the canonical raw format).
+3. Converts the CSV corpus into 4 target storage formats: JSON-LD, SQLite (optimized), PostgreSQL (indexed + partitioned), RDF TripleStore.
+4. Benchmarks 7 laboratory-standard retrieval scenarios (Q1-Q7) against all 5 formats, measuring wall time, peak RAM, CPU utilization, disk I/O, and disk footprint.
+5. Extrapolates measured results to 600 GB and beyond (0.6 / 1.2 / 6 TB) using complexity-aware scaling models.
+6. Produces summary tables and charts that clearly rank storage formats for laboratory use (analysis, reporting, search/filter, archiving, inter-lab exchange).
+
+## Execution environment
+- Claude Code agent inside Cursor, local filesystem access, on-premise only (no cloud services).
+- Python 3.11+ (tabs for indentation), PostgreSQL 16 local instance, SQLite via stdlib `sqlite3`, `rdflib` for RDF, optional local triplestore (Jena Fuseki / GraphDB Free) - if unavailable, use `rdflib` + on-disk `oxigraph` (pip) as the queryable store.
+- All artifacts written under `./out/` relative to repo root:
+ - `./out/csv/` `./out/json/` `./out/sqlite/` `./out/pg/` (dumps) `./out/rdf/` `./out/bench/` `./out/report/`
+
+## Task files and dependencies
+```mermaid
+flowchart TD
+    T01[01_lab_configuration] --> T02[02_process_flow_diagrams]
+    T01 --> T03[03_csv_data_generation]
+    T03 --> T04[04_convert_json]
+    T03 --> T05[05_convert_sqlite]
+    T03 --> T06[06_convert_postgresql]
+    T03 --> T07[07_convert_rdf]
+    T04 --> T08[08_benchmark_queries]
+    T05 --> T08
+    T06 --> T08
+    T07 --> T08
+    T03 --> T08
+    T08 --> T09[09_benchmark_execution]
+    T09 --> T10[10_extrapolation_model]
+    T09 --> T11[11_reporting]
+    T10 --> T11
+```
+
+## Parallelism
+- T04-T07 are independent; run in parallel after T03.
+- T02 is independent of data generation; can run any time after T01.
+
+## Execution order
+01 → 02 → 03 → (04 | 05 | 06 | 07) → 08 → 09 → 10 → 11
+
+## Global conventions
+- IDs: `batch:B721`, `wafer:B721-W1`, `coupon:B721-W1-C07`, `track:B721-W1-C07-T2`, run IDs `run:R1..R4`.
+- All timestamps ISO-8601 UTC. All units SI-explicit in column names (`_mN`, `_um`, `_GPa`, `_wtpct`, `_nm`).
+- Random seed fixed at 20260711 for full reproducibility.
+- Python indentation: tabs.
+- Every task writes a completion marker `./out/.done/<task>.ok` containing row/byte counts for downstream validation.

docs/specs/01_lab_configuration.md

@@ -0,0 +1,56 @@
+# 01 - Laboratory Configuration
+
+## Context / Goal
+Define the authoritative laboratory model that all downstream tasks (data generation, ontology, schemas) must follow. This file is a specification, not code; produce `./out/config/lab_config.yaml` capturing everything below.
+
+## Dependencies
+None (root task).
+
+## Material system
+- Substrate: Ti-6Al-4V coupons, 10 × 10 × 3 mm.
+- Adhesion layer: Cr (sputtered).
+- Functional coating: Pt-Au deposited in a **composition gradient** across each wafer; film thickness range 0.3-1.1 µm.
+
+## Instruments
+| Role | Instrument | Data produced |
+|---|---|---|
+| Friction | RAPID custom high-throughput parallelized 6-probe tribometer | COF vs cycle per track; test conditions |
+| Nanoindentation | Bruker TI980 TriboIndenter | Hardness, reduced modulus (25 indents/coupon) |
+| Composition | Bruker M4 Tornado micro-XRF (assumed model) | Pt/Au wt% map per coupon (20×20 grid) |
+| Deposition | Kurt J. Lesker PVD 200 sputter-down | Batch deposition parameters |
+| Surface roughness | AFM | Topography image metadata + Ra/Rq per coupon |
+| Film thickness | Optical profilometry | Thickness map per coupon (10×10 grid) |
+| Simulation | SIMTRA (sputter transport Monte-Carlo) | Deposition atom-energy / composition profiles per deposition run |
+
+## Deposition matrix (4 batches)
+| Batch | Pt:Au gun tilt | Pt power | Au power | Pt discharge | Au discharge |
+|---|---|---|---|---|---|
+| B721 | 20°:0°  | 150 W | 50 W  | 432 V | 311 V |
+| B722 | 20°:20° | 100 W | 100 W | 399 V | 352 V |
+| B723 | 20°:20° | 150 W | 50 W  | 430 V | 311 V |
+| B724 | 0°:20°  | 50 W  | 150 W | 376 V | 340 V |
+
+## Hierarchy and volumes
+```
+batch (×4) → wafer (×3 per batch) → coupon (×49 per wafer, 7×7 grid) = 588 coupons
+```
+- Friction testing: 480 coupons (4 RAPID runs × 120 coupons: 5 plates × 6 probes × 4 coupons/square).
+- Reserve/QA: 108 coupons (no friction data; characterization only).
+- Environments: run R1,R2 = Lab Air (µ_normal); run R3,R4 = Dry N2 (µ_dry_nit).
+- Tracks: 3 replicate tracks per friction coupon, identical conditions, fresh counterface ball per track (3 balls per holder) → 1,440 tracks.
+- Cycles: 1,000 reciprocating cycles per track → 1,440,000 cycle records.
+- Friction loops: every 100th cycle stores a position-resolved loop of 200 points → 1,440 × 10 × 200 = 2,880,000 loop points.
+- Nominal test conditions: normal load 100 mN, stroke 1 mm, sliding speed 1 mm/s, ruby (Al2O3) ball Ø 3.175 mm; RH 45% (Lab Air) / 2% (Dry N2); T 23 ± 1 °C.
+
+## Physical models for value simulation (used in task 03)
+- Coupon Au wt%: linear gradient across wafer x-position, batch-dependent center: B721 8%, B722 25%, B723 10%, B724 60%; ±5 wt% span across wafer; XRF grid adds N(0, 0.3) noise.
+- COF run-in: `cof(c) = cof_ss + (cof_0 - cof_ss)·exp(-c/τ) + N(0, σ)`; cof_0 ∈ [0.35, 0.5]; τ ∈ [100, 400] cycles; σ = 0.01.
+- Steady-state COF depends on Au% and environment: Lab Air `cof_ss = 0.32 - 0.0015·Au%`; Dry N2 `cof_ss = 0.24 - 0.0012·Au%`; clamp ≥ 0.12.
+- Hardness: `H = 8.5 - 0.05·Au% + N(0, 0.25)` GPa; reduced modulus `Er = 190 - 0.6·Au% + N(0, 4)` GPa.
+- Roughness Ra: log-normal, median 5 nm, σ_log 0.3.
+- Thickness: 0.3-1.1 µm, radial wafer profile + noise.
+- Wear volume per track: Archard-like `V = k·F·s`, k depends on Au% and environment; store as track-level derived measurement.
+
+## Output
+- `./out/config/lab_config.yaml` - machine-readable version of everything above (volumes, matrices, model coefficients, seed).
+- Marker `./out/.done/01.ok`.

docs/specs/02_process_flow_diagrams.md

@@ -0,0 +1,48 @@
+# 02 - Process Flow Diagrams
+
+## Context / Goal
+Generate the comprehensive Mermaid documentation of the laboratory workflow (reconstructed from the 10 Miro slides) as standalone `.md` / `.mermaid` artifacts for the final report and for human review.
+
+## Dependencies
+01 (uses volumes/terminology from lab_config.yaml).
+
+## Task
+Produce `./out/report/process_flow.md` containing the following diagrams, each also saved as a separate `.mermaid` file in `./out/report/diagrams/`.
+
+### D1 - Coupon Assembly
+```mermaid
+flowchart LR
+    A([Ti-6Al-4V Base 10x10x3 mm]) --> B[Cleaning]
+    B --> B1([Rinsed in deionized water])
+    B1 --> B2([Sonicated in cleaning solution])
+    B --> C[Smearing Adhesive]
+    C --> C1([Cr adhesive coating])
+    C --> D[Vapor Deposition System<br/>Kurt J. Lesker PVD 200]
+    D --> D1([PtAu sputter coating, gradient])
+    D --> E([Test Coupon])
+```
+
+### D2 - Characterization & Batch Assembly
+Test Coupon → SIMTRA (composition / atom energies) → Optical profilometry (film thickness 0.3-1.1 µm) → assemble Test Wafer (×49 coupons) → assemble Test Batch (×3 wafers) → batches 721-724 with the deposition parameter table.
+
+### D3 - Testing tree
+Friction (RAPID; Lab Air → µ_normal; Dry N2 → µ_dry_nit), Nanoindentation (TI980 → hardness, reduced modulus), AFM (topography → Ra), XRF (M4 Tornado → Pt/Au map).
+
+### D4 - Tribometer session sequence (sequenceDiagram)
+Request test → Get samples → Load samples → Load ball holders (3 counterfaces/holder) → Create Excel test plan (Sample Plate, Plate Location, Sample ID, Save Location, Folder Name, X/Y/Z Offset, Load, Iterations) → Transfer to tribometer → Set up platter → Activate software → Load plan → Activate tribometer → ... → Software stop → Save avg files → Remove plates/platters → Close software → Pull equipment out.
+
+### D5 - Execution loop (flowchart)
+```
+Load all counter-faces
+  FOR EACH of 5 plates
+    FOR EACH of 6 probes (parallel)
+      FOR EACH of 4 coupons on coupon square
+        3x per coupon: draw track -> rotate counter-face
+```
+
+### D6 - Data hierarchy ERD-style
+batch → wafer → coupon → {xrf_map, nanoindentation, afm, profilometry} and coupon → track → {cycles, loops, wear}; deposition → simtra_profile.
+
+## Output
+- `./out/report/process_flow.md`, `./out/report/diagrams/D1..D6.mermaid`
+- Marker `./out/.done/02.ok`.

docs/specs/03_csv_data_generation.md

@@ -0,0 +1,46 @@
+# 03 - Simulated CSV Data Generation (~150 MB)
+
+## Context / Goal
+Generate the canonical raw dataset as a hierarchical CSV file tree, exactly as instruments would write it. This corpus is the single source for all conversions and benchmarks. Target size: 140-170 MB.
+
+## Dependencies
+01 (`lab_config.yaml`).
+
+## Input
+`./out/config/lab_config.yaml` - volumes, deposition matrix, physical model coefficients, seed 20260711.
+
+## Processing instructions
+Write `generate_data.py` (tabs for indentation, pure stdlib + numpy). Stream rows; never hold the full cycle dataset in memory. Use `random.Random(seed)` / `numpy.random.default_rng(seed)` only.
+
+### File tree to produce
+```
+out/csv/
+├── batches.csv                      # 4 rows: deposition params
+├── simtra/
+│   └── simtra_profile_<batch>.csv   # 4 files, 1000 rows each: angle_deg, energy_eV, pt_flux, au_flux
+├── batch_<B>/wafer_<W>/
+│   ├── wafer_info.csv
+│   └── coupon_<C>/
+│       ├── coupon_info.csv          # position on wafer, thickness_um, ra_nm, assigned run/plate/probe/square or RESERVE
+│       ├── xrf_map.csv              # 400 rows: grid_x, grid_y, pt_wtpct, au_wtpct
+│       ├── profilometry.csv         # 100 rows: grid_x, grid_y, thickness_um
+│       ├── nanoindentation.csv      # 25 rows: indent_id, x_um, y_um, hardness_GPa, reduced_modulus_GPa, max_load_mN
+│       ├── afm.csv                  # 1 row: ra_nm, rq_nm, image_file (synthetic path)
+│       └── track_<T>/               # only for 480 friction coupons, T in 1..3
+│           ├── track_info.csv       # run_id, environment, load_mN, stroke_mm, speed_mm_s, counterface_id, started_at
+│           ├── cof_vs_cycle.csv     # 1000 rows: cycle, cof
+│           ├── friction_loops.csv   # 2000 rows: cycle (every 100th), position_um (200 pts), friction_force_mN
+│           └── wear.csv             # 1 row: wear_volume_um3, k_archard, sliding_distance_m
+└── runs.csv                         # 4 rows: run_id, environment, date, plates, operator
+```
+
+### Value generation
+Apply exactly the physical models of task 01 (gradient Au%, run-in COF, hardness/modulus vs Au%, Archard wear). COF must correlate with the coupon's mean Au% and the run environment so that Q2/Q3/Q7 benchmarks return physically meaningful results.
+
+### Size control
+Expected: cycles 1.44 M rows (~30 MB) + loops 2.88 M rows (~95 MB) + XRF 235 K rows (~8 MB) + remainder (~10 MB) ≈ 150 MB. After generation, print actual tree size; abort with error if outside 120-200 MB.
+
+## Output
+- Full tree under `./out/csv/` (~150 MB).
+- `./out/csv/MANIFEST.csv`: file path, rows, bytes, sha256 - used by conversion tasks for validation.
+- Marker `./out/.done/03.ok` with total rows/bytes.

docs/specs/04_convert_json.md

@@ -0,0 +1,22 @@
+# 04 - Convert to JSON-LD
+
+## Context / Goal
+Produce the ontology-annotated JSON-LD representation of the corpus: the format evaluated for inter-lab data exchange (FAIR). Two variants: (a) full - bulk points embedded; (b) hybrid - metadata JSON-LD with `sourceFile` links to CSV.
+
+## Dependencies
+03 (`./out/csv/`, MANIFEST.csv).
+
+## Input
+CSV tree.
+
+## Processing instructions
+- `@context`: project vocab `https://sandia.gov/ontology/tribology#` + QUDT units + PROV-O provenance; predicates `partOf`, `cutFrom`, `performedOn`, `performedBy` as `@id` links.
+- Entity classes: `Batch, Wafer, Coupon, Instrument, CompositionMeasurement, NanoindentationMeasurement, AFMMeasurement, ProfilometryMeasurement, FrictionTest (track), FrictionCycle, FrictionLoopPoint, WearMeasurement, SimtraProfile, Run`.
+- Variant FULL: one JSON-LD file per coupon (embedding its tracks/cycles/loops), compact serialization (no pretty-print). Directory `./out/json/full/`.
+- Variant HYBRID: single `./out/json/hybrid/dataset.jsonld` with all metadata entities; cycle/loop arrays replaced by `{"sourceFile": "<relative csv path>", "rows": N, "sha256": "..."}`.
+- Stream-write; validate one sample file with `json.load` + rdflib JSON-LD parse.
+
+## Output
+- `./out/json/full/` (expect ~0.7-1.2 GB), `./out/json/hybrid/dataset.jsonld` (expect < 20 MB).
+- Record byte sizes of both variants into `./out/bench/storage_sizes.csv` (append: format, variant, bytes).
+- Marker `./out/.done/04.ok`.

docs/specs/05_convert_sqlite.md

@@ -0,0 +1,32 @@
+# 05 - Convert to SQLite (optimized)
+
+## Context / Goal
+Build the single-file relational variant optimized for size and single-user query speed.
+
+## Dependencies
+03.
+
+## Input
+CSV tree.
+
+## Processing instructions
+- One database `./out/sqlite/tribo.db`. PRAGMAs during load: `journal_mode=OFF, synchronous=OFF, cache_size=-262144, foreign_keys=ON`; after load: `journal_mode=WAL`, `ANALYZE`, `VACUUM`.
+- Foreign keys are enforced on every table, including bulk tables (variant B of the measured key-schema study; owner decision 2026-07-11, see db-sql-schema rule). Full study with ER diagrams and measurements: [../research/Tribology_FK_Architecture.html](../research/Tribology_FK_Architecture.html).
+- Schema (16-table style, integer surrogate keys everywhere; no TEXT foreign keys in bulk tables; SQL column names are the lowercase of the CSV headers per data-naming-units):
+	- `batches, wafers, coupons, runs, instruments, tracks` - dimension tables, TEXT natural keys kept as unique columns.
+	- `simtra_profiles(batch_id INT, row_no INT, angle_deg REAL, energy_ev REAL, pt_flux REAL, au_flux REAL, PRIMARY KEY(batch_id, row_no)) WITHOUT ROWID`.
+	- `xrf_points(coupon_id INT, grid_x INT, grid_y INT, pt_wtpct REAL, au_wtpct REAL, PRIMARY KEY(coupon_id, grid_x, grid_y)) WITHOUT ROWID`.
+	- `profilometry_points(coupon_id INT, grid_x INT, grid_y INT, thickness_um REAL, PRIMARY KEY(coupon_id, grid_x, grid_y)) WITHOUT ROWID`.
+	- `nanoindentation(coupon_id INT, indent_id INT, x_um REAL, y_um REAL, hardness_gpa REAL, reduced_modulus_gpa REAL, max_load_mn REAL, PRIMARY KEY(coupon_id, indent_id)) WITHOUT ROWID`.
+	- `afm(coupon_id INT PRIMARY KEY, ra_nm REAL, rq_nm REAL, image_file TEXT)`.
+	- `friction_cycles(track_id INT, cycle INT, cof REAL, PRIMARY KEY(track_id, cycle)) WITHOUT ROWID`.
+	- `friction_loop_points(track_id INT, cycle INT, pt INT, position_um REAL, friction_force_mn REAL, PRIMARY KEY(track_id, cycle, pt)) WITHOUT ROWID`.
+	- `wear(track_id INT PRIMARY KEY, wear_volume_um3 REAL, k_archard REAL, sliding_distance_m REAL)`.
+	- `track_summary(track_id INT PRIMARY KEY, cof_ss_mean REAL, cof_ss_std REAL, run_in_cycles INT)` - precomputed during load (algorithm fixed in db-sql-schema rule, section 5).
+- Indexes for Q1-Q7: `coupons(au_wtpct_mean)`, `tracks(run_id)`, `tracks(coupon_id)`, `runs(environment)`, composite `tracks(environment, load_mn)` (denormalize environment/load onto tracks).
+- Batch inserts with `executemany`, 50k rows per transaction.
+- Validate row counts against MANIFEST.
+
+## Output
+- `./out/sqlite/tribo.db`; size appended to `storage_sizes.csv`.
+- Marker `./out/.done/05.ok`.

docs/specs/06_convert_postgresql.md

@@ -0,0 +1,23 @@
+# 06 - Convert to PostgreSQL (indexed + partitioned)
+
+## Context / Goal
+Build the production-grade relational variant: the system-of-record candidate per the existing Storage ADR (PostgreSQL primary, RDF as deferred virtual layer).
+
+## Dependencies
+03.
+
+## Input
+CSV tree.
+
+## Processing instructions
+- Local PostgreSQL 16, database `tribo`. Same logical schema as task 05 with PostgreSQL types (`INT`, `REAL`, `TIMESTAMPTZ`, `TEXT`).
+- `friction_cycles` and `friction_loop_points`: declarative RANGE partitioning by `track_id` (12 partitions of 120 tracks) - models the partitioning strategy needed at 600 GB scale.
+- Load via `COPY FROM STDIN` (psycopg3 `copy`), not INSERT.
+- Indexes after load: PK b-trees; `tracks(environment, load_mn)`, `coupons(au_wtpct_mean)`, `nanoindentation(coupon_id)`, BRIN on `friction_cycles(cycle)` per partition.
+- `track_summary` as a materialized view (`cof_ss_mean` = avg of cycles > run_in threshold; `run_in_cycles` via first cycle where |cof - cof_ss| < 2σ).
+- `VACUUM ANALYZE` after load. Record `pg_total_relation_size` per table and total into `storage_sizes.csv`.
+- Also export a compressed dump `./out/pg/tribo.dump` (pg_dump -Fc) for archiving-format comparison.
+
+## Output
+- Live `tribo` database + `./out/pg/tribo.dump`.
+- Marker `./out/.done/06.ok`.

docs/specs/07_convert_rdf.md

@@ -0,0 +1,24 @@
+# 07 - Convert to RDF TripleStore
+
+## Context / Goal
+Build the semantic-web variant to quantify triple-store overhead and SPARQL performance on real hardware.
+
+## Dependencies
+03 (optionally reuse 04 hybrid JSON-LD as an intermediate).
+
+## Input
+CSV tree (or `./out/json/full/` for direct rdflib parsing - choose the faster path, document the choice).
+
+## Processing instructions
+- Vocabulary identical to task 04 (`tribology#` + QUDT + PROV-O).
+- Triple modeling: every cycle → 3 triples (`rdf:type` optional at bulk level to save space - model cycles as `track :hasCycle [ :cycleNumber n ; :cof x ]` without explicit type; document the decision). Loop points likewise. Expected total: 25-45 M triples.
+- Serialize `./out/rdf/dataset.nt.gz` (N-Triples, gzip) and `./out/rdf/dataset.ttl` (Turtle) - both sizes recorded.
+- Queryable store, first available option:
+	1. `oxigraph` (pip, embedded, on-disk store `./out/rdf/oxigraph_store/`) - preferred, no external service.
+	2. Jena Fuseki with TDB2 if Java present.
+- Bulk-load, record store on-disk size into `storage_sizes.csv` and load wall-time.
+- Smoke-test with a SPARQL COUNT query.
+
+## Output
+- Serializations + populated store.
+- Marker `./out/.done/07.ok` with triple count.

docs/specs/08_benchmark_queries.md

@@ -0,0 +1,31 @@
+# 08 - Benchmark Query Definitions (Q1-Q7 × 5 formats)
+
+## Context / Goal
+Define the 7 laboratory-standard retrieval scenarios and their concrete implementation for each of the 5 formats. Every implementation must return identical result sets (validated by checksum of sorted result rows, tolerance 1e-9 on floats).
+
+## Dependencies
+04, 05, 06, 07 (all stores populated) + 03 (CSV baseline).
+
+## Scenarios
+| ID | Scenario | Result shape |
+|---|---|---|
+| Q1 | COF vs cycle curve for one track (`B722-W2-C13-T2`) | 1,000 rows (cycle, cof) |
+| Q2 | Steady-state COF for all coupons with mean Au = 10 ± 0.5 wt% | ~dozens of rows (coupon, env, cof_ss) |
+| Q3 | Hardness vs steady-state COF across all batches (join nano × track_summary) | 480 rows |
+| Q4 | Tracks in Dry N2, load 100 mN, cof_ss > 0.20 (anomaly filter) | filtered rows |
+| Q5 | Mean run-in cycles per batch (aggregation over ALL cycle rows, no summary tables allowed - force the full scan) | 4 rows |
+| Q6 | Wear volume vs load per coupon (join wear × tracks × coupons) | 480 rows |
+| Q7 | Stribeck-style aggregation: mean COF grouped by (environment, speed·load bucket) over all cycles past run-in | ~10 rows |
+
+## Implementations
+- **CSV**: pure Python + csv module streaming (pandas allowed only as a second measured variant, labeled separately). Q1 = direct path read (this is the honest CSV strength). Q2-Q7 = directory walks + programmatic joins.
+- **JSON**: variant FULL, `ijson` streaming parser; Q1 = load the single coupon file.
+- **SQLite**: SQL over `tribo.db`, cold and warm cache runs (`PRAGMA cache_size` default; drop OS cache not required, just report both run 1 and run 3).
+- **PostgreSQL**: SQL, `EXPLAIN (ANALYZE, BUFFERS)` captured alongside; parallel workers default.
+- **RDF**: SPARQL against the populated store; capture query text.
+- Q5 rule applies to all formats: query raw cycle data, not `track_summary`.
+
+## Output
+- `./out/bench/queries/` - one subfolder per format with 7 runnable query files (`q1.sql`, `q1.sparql`, `q1.py`, ...).
+- `./out/bench/expected/q<N>.sha256` - canonical result checksums (computed once from PostgreSQL, cross-validated against SQLite).
+- Marker `./out/.done/08.ok`.

docs/specs/09_benchmark_execution.md

@@ -0,0 +1,22 @@
+# 09 - Benchmark Execution & Resource Measurement
+
+## Context / Goal
+Execute all 35 (Q1-Q7 × 5) benchmark cells with rigorous resource metering; produce the raw measurement table that feeds extrapolation and reporting.
+
+## Dependencies
+08.
+
+## Processing instructions
+Write `bench_runner.py` (tabs):
+- Each cell runs as a **separate subprocess** so RSS/CPU are attributable. For PostgreSQL, also sample the backend PID via `pg_stat_activity` + psutil.
+- Metrics per run: wall time (s), peak RSS (MB, psutil `memory_info().rss` sampled at 50 ms), CPU time user+sys (s), CPU utilization % (cpu_time / wall / cores), read bytes (psutil `io_counters`), result row count, checksum-match flag.
+- Protocol: 1 warm-up run (discarded) + 3 measured runs per cell; report median and min/max. Randomize cell order to spread cache effects; record run order.
+- Cold-cache variant: repeat the full matrix once after dropping caches if permitted (`sync; echo 3 > /proc/sys/vm/drop_caches` requires root - if not available, skip and note it).
+- Failures/timeouts: hard timeout 30 min per run; record `TIMEOUT` and continue (expected possibility: RDF Q5).
+- Storage footprint: consolidate `storage_sizes.csv` (all formats, incl. gzip-compressed archival sizes of each: `tar.gz` of CSV tree, gzip JSON, sqlite db gz, pg dump, nt.gz).
+
+## Output
+- `./out/bench/results_raw.csv`: format, query, run, wall_s, peak_rss_mb, cpu_s, cpu_util_pct, read_mb, rows, checksum_ok.
+- `./out/bench/results_median.csv` - aggregated.
+- `./out/bench/storage_sizes.csv` - final consolidated.
+- Marker `./out/.done/09.ok`.

docs/specs/10_extrapolation_model.md

@@ -0,0 +1,38 @@
+# 10 - Extrapolation to 600 GB and Beyond
+
+## Context / Goal
+Project the 150 MB measurements to production scale: 600 GB (×4000), 1.2 TB (×8000), 6 TB (×40000) of raw CSV. Model both storage growth and query-performance degradation to understand system requirements as laboratory data grows.
+
+## Dependencies
+09 (`results_median.csv`, `storage_sizes.csv`).
+
+## Processing instructions
+Write `extrapolate.py`:
+
+### Storage scaling
+Linear in data volume: `size_at(S) = measured_size × S / S_measured`, per format (CSV 1×; report the measured coefficients for JSON, SQLite, PG incl. index share, RDF store).
+
+### Query-time scaling laws (apply per format × query)
+| Access pattern | Law | Applies to |
+|---|---|---|
+| Direct path / keyed point read | O(1) / O(log n) | CSV Q1, SQLite/PG Q1, RDF Q1 |
+| Indexed selective filter | O(log n + k) | SQLite/PG Q2, Q4, Q6; RDF Q2, Q4 |
+| Indexed join over dimensions | O(k log n) | Q3, Q6 relational; RDF joins ×(2-5) constant penalty |
+| Full scan / aggregation | O(n) | all CSV/JSON Q2-Q7; Q5 and Q7 everywhere |
+| Single-threaded full scan | O(n), 1 core | SQLite Q5/Q7 |
+| Parallel partitioned scan | O(n / min(cores, partitions)) | PostgreSQL Q5/Q7 |
+
+Calibrate the constant from the measured 150 MB point; project wall time at each scale. Flag any projection > 1 h as "IMPRACTICAL", > 24 h as "FAIL".
+
+### RAM model
+- CSV/JSON: streaming - flat RAM; note that pandas variant would be O(n) and mark it infeasible > ~10 GB.
+- SQLite/PG: recommended cache = hot-set model: 10% of index size + working set; report per scale.
+- RDF: dictionary + index working set ≈ 15-25% of store size for acceptable latency; report where it exceeds 1 TB RAM (single-node infeasibility threshold).
+
+### Hardware sizing table
+For each format × scale: required disk (with 30% headroom), recommended RAM, recommended cores; approximate cost using parameters in `./out/config/hw_prices.yaml` (defaults, editable): DDR5 RDIMM $14/GB, enterprise NVMe $250/TB, base 16-core 1U $6k, 32-core $10k, per extra node $6k.
+
+## Output
+- `./out/bench/extrapolation.csv`: format, query, scale_gb, projected_wall_s, flag.
+- `./out/bench/hardware_sizing.csv`: format, scale_gb, disk_tb, ram_gb, cores, nodes, est_cost_usd.
+- Marker `./out/.done/10.ok`.

docs/specs/11_reporting.md

@@ -0,0 +1,37 @@
+# 11 - Reporting: Tables, Charts, Weighted Scoring
+
+## Context / Goal
+Aggregate all measurements and projections into the final decision document that clearly ranks the storage formats for laboratory use cases: analysis, report generation, search/filtering, archiving, inter-lab exchange.
+
+## Dependencies
+09, 10 (and 02 for embedding process diagrams).
+
+## Processing instructions
+Write `report.py` using matplotlib (PNG, 150 dpi, log scales where ranges span decades) into `./out/report/charts/`, and assemble `./out/report/REPORT.md`.
+
+### Tables
+1. Storage footprint: measured @150 MB + coefficients vs CSV + projected @600 GB/1.2 TB/6 TB (incl. compressed archival sizes).
+2. Q1-Q7 median wall time / peak RAM / CPU% / read MB, per format (measured).
+3. Q1-Q7 projected wall time @600 GB with IMPRACTICAL/FAIL flags.
+4. Hardware sizing & cost per format @600 GB.
+5. Weighted scoring matrix (see below).
+
+### Charts (each measured + each projected @600 GB)
+- C1: disk footprint bar (log).
+- C2: RAM requirement bar (log).
+- C3: per-query grouped bars Q1-Q7 wall time (log), one chart per query, 5 formats each.
+- C4: degradation curves - wall time vs data size (150 MB → 6 TB), one line per format, per query class (point read / indexed / full aggregation), log-log.
+- C5: stacked-bar weighted score.
+- C6: cost vs performance scatter @600 GB (x = est cost, y = geometric-mean query time, log-log).
+
+### Weighted scoring (agreed methodology)
+Normalized 0-10 per metric, then weights: Search speed ×5 (0-50), RAM economy ×2 (0-20), Disk economy ×1 (0-10). Max 80. Search score = normalized inverse geometric mean of Q1-Q7 projected @600 GB (TIMEOUT/FAIL → 0). Present measured-scale and 600 GB-scale scoreboards side by side.
+
+### Narrative sections (concise, English)
+- Executive summary with the recommendation (expected, to be confirmed by data: PostgreSQL system of record; CSV retained as raw archive; hybrid JSON-LD for exchange; RDF as virtual layer only - reference Ontop OBDA).
+- Use-case mapping table: analysis / reports / search / archiving / exchange → recommended format(s).
+- Limitations: single-node, simulated data, extrapolation assumptions.
+
+## Output
+- `./out/report/REPORT.md` + `./out/report/charts/*.png` + all tables as CSV in `./out/report/tables/`.
+- Marker `./out/.done/11.ok`.

extrapolate.py

@@ -0,0 +1,295 @@
+"""Task 10 - Extrapolation to 600 GB / 1.2 TB / 6 TB.
+
+Calibrates per-cell scaling laws on the measured 150 MB point (task 09
+medians) and projects wall times and hardware needs per format x scale.
+Model (spec 10, docs/rules/bench-methodology.md section 7):
+- per-format fixed overhead = the format's cheapest measured median
+  (interpreter start, imports, connection); the remainder is the
+  data-dependent calibration constant;
+- growth laws per access pattern: o1 (flat), log (index depth,
+  1 + ln r / ln n0 with n0 = measured friction-cycle rows), logk (result
+  set grows linearly with data), k_log_n (linear x index depth), scan
+  (linear), scan_parallel (linear x base_cores / cores_at_scale; range
+  partitions outnumber cores at every scale);
+- RDF join penalties are already embedded in the measured constants;
+- RDF q3/q5/q7 are scans because the implementation derives run-in /
+  steady-state from raw cycles (no summary triples exist by design).
+Every number written here is a PROJECTION, never a measurement; rows are
+flagged IMPRACTICAL (> 1 h) and FAIL (> 24 h).
+Spec: docs/specs/10_extrapolation_model.md.
+"""
+
+from __future__ import annotations
+
+import csv
+import logging
+import math
+import sys
+from pathlib import Path
+
+import psycopg
+import yaml
+
+from common.pg import get_dsn
+from common.pipeline import (
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	remove_stale_marker,
+	write_marker,
+)
+from common.relational import expected_counts
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "10"
+DEPENDS_ON = ["09"]
+
+SCALES_GB = (600, 1200, 6000)  # raw-CSV target scales (decimal GB), spec 10
+IMPRACTICAL_S = 3600.0         # projection > 1 h, spec 10
+FAIL_S = 24 * 3600.0           # projection > 24 h, spec 10
+DISK_HEADROOM = 1.3            # 30% free-space headroom, spec 10
+CACHE_INDEX_FRACTION = 0.10    # relational hot-set: 10% of index size, spec 10
+WORKING_SET_GB = 4.0           # relational session working set, spec 10 default
+RDF_HOT_FRACTION = 0.20        # dictionary+index working set, 15-25% midpoint, spec 10
+STREAM_RAM_GB = 16.0           # flat streaming baseline (csv/json servers)
+MIN_RAM_GB = 16.0              # smallest sensible server RAM
+NODE_RAM_CEILING_GB = 1024.0   # single-node RAM threshold, spec 10 RDF note
+BASE_CORES = 4                 # cores behind the measured PG parallel scans (task 09 host)
+CORE_OPTIONS = (16, 32)        # priced server tiers, hw_prices.yaml
+
+FORMATS = ("csv", "json", "sqlite", "pg", "rdf")
+QUERY_NUMBERS = tuple(range(1, 8))
+
+# working (queryable) footprint variant per format for storage scaling
+SIZE_VARIANT = {
+	"csv": ("csv", "tree"), "json": ("json", "full"), "sqlite": ("sqlite", "db"),
+	"pg": ("pg", "live"), "rdf": ("rdf", "store"),
+}
+
+# access-pattern law per (format, query) - spec 10 table, adjusted to the
+# ACTUAL implementations (rdf q3/q5/q7 read raw cycles; see module docstring)
+LAWS = {
+	("csv", 1): "o1", ("json", 1): "o1", ("sqlite", 1): "log", ("pg", 1): "log", ("rdf", 1): "log",
+	("csv", 2): "scan", ("json", 2): "scan", ("sqlite", 2): "logk", ("pg", 2): "logk", ("rdf", 2): "logk",
+	("csv", 3): "scan", ("json", 3): "scan", ("sqlite", 3): "k_log_n", ("pg", 3): "k_log_n", ("rdf", 3): "scan",
+	("csv", 4): "scan", ("json", 4): "scan", ("sqlite", 4): "logk", ("pg", 4): "logk", ("rdf", 4): "logk",
+	("csv", 5): "scan", ("json", 5): "scan", ("sqlite", 5): "scan", ("pg", 5): "scan_parallel", ("rdf", 5): "scan",
+	("csv", 6): "scan", ("json", 6): "scan", ("sqlite", 6): "k_log_n", ("pg", 6): "k_log_n", ("rdf", 6): "k_log_n",
+	("csv", 7): "scan", ("json", 7): "scan", ("sqlite", 7): "scan", ("pg", 7): "scan_parallel", ("rdf", 7): "scan",
+}
+
+HW_PRICES_DEFAULT = """# Hardware cost parameters for extrapolation (task 10).
+# Street prices retrieved 2026-07-11 from the sources below (server DRAM is
+# in a documented 2025-2026 price surge - see market_context). The cost
+# model buys whole components: 64 GB RDIMM modules and 7.68 TB NVMe drives.
+# Operator-editable; re-run extrapolate.py after changes (code-config-yaml).
+as_of: "2026-07-11"
+ram_module_gb: 64
+ram_module_usd: 1887.0       # A-Tech 64GB (2x32GB) DDR5-5600 ECC RDIMM
+nvme_drive_tb: 7.68
+nvme_drive_usd: 3995.0       # Cloud Ninjas NEW 7.68TB NVMe U.2 1DWPD (Dell 14-16G)
+server_16_core_usd: 4618.0   # Supermicro AS-1015CS-TNR 1U (EPYC 9004/9005), Broadberry starting config
+server_32_core_usd: 6272.0   # Supermicro AS-1115CS-TNR 1U, Broadberry starting config
+extra_node_usd: 4618.0       # one additional entry 1U node (AS-1015CS-TNR chassis)
+sources:
+  ram: https://atechmemory.com/collections/ddr5-memory-ram
+  nvme: https://cloudninjas.com/products/new-7-68tb-nvme-u-2-1dwpd-sie-2-5-enterprise-solid-state-drive-for-14th-15th-16th-gen-dell
+  servers: https://www.broadberry.com/amd-epyc-9004-supermicro-servers
+  market_context: https://www.techpowerup.com/342331/server-dram-pricing-jumps-50-only-70-of-orders-getting-filled
+"""
+HW_PRICE_KEYS = (
+	"as_of", "ram_module_gb", "ram_module_usd", "nvme_drive_tb", "nvme_drive_usd",
+	"server_16_core_usd", "server_32_core_usd", "extra_node_usd", "sources",
+)
+
+
+def read_medians(path: Path) -> dict[tuple[str, int], float]:
+	"""Median walls of the 35 cells; abort unless every cell is OK + checksummed
+	(projections built on invalid measurements are forbidden, bench-methodology).
+	"""
+	medians: dict[tuple[str, int], float] = {}
+	with open(path, encoding="ascii", newline="") as fh:
+		for row in csv.DictReader(fh):
+			if row["status"] != "OK" or row["checksum_ok"] != "true":
+				raise ValueError(f"{path}: cell {row['format']}/{row['query']} is {row['status']}/checksum={row['checksum_ok']} - cannot calibrate on it")
+			medians[(row["format"], int(row["query"][1:]))] = float(row["wall_s_median"])
+	expected_cells = {(fmt, qn) for fmt in FORMATS for qn in QUERY_NUMBERS}
+	if set(medians) != expected_cells:
+		raise ValueError(f"{path}: {len(medians)} cells, expected the full 35-cell matrix")
+	return medians
+
+
+def read_sizes(path: Path) -> dict[tuple[str, str], int]:
+	sizes: dict[tuple[str, str], int] = {}
+	with open(path, encoding="ascii", newline="") as fh:
+		reader = csv.reader(fh)
+		next(reader)
+		for fmt, variant, nbytes in reader:
+			sizes[(fmt, variant)] = int(nbytes)
+	missing = set(SIZE_VARIANT.values()) - set(sizes)
+	if missing:
+		raise ValueError(f"{path}: missing rows {sorted(missing)}")
+	return sizes
+
+
+def pg_index_share(dsn: str) -> tuple[float, int]:
+	"""Index share of the live PG footprint (read-only catalog query)."""
+	with psycopg.connect(dsn) as conn:
+		index_bytes, total_bytes = conn.execute(
+			"SELECT COALESCE(sum(pg_indexes_size(c.oid)), 0)::bigint,"
+			"       COALESCE(sum(pg_total_relation_size(c.oid)), 0)::bigint"
+			" FROM pg_class c JOIN pg_namespace n ON n.oid = c.relnamespace"
+			" WHERE n.nspname = 'public' AND c.relkind IN ('r', 'p', 'm')"
+		).fetchone()
+	if total_bytes == 0:
+		raise ValueError("pg catalog reports zero relation bytes - is the database loaded?")
+	return index_bytes / total_bytes, index_bytes
+
+
+def load_hw_prices(config_dir: Path) -> dict:
+	path = config_dir / "hw_prices.yaml"
+	if not path.exists():
+		path.write_text(HW_PRICES_DEFAULT, encoding="ascii", newline="\n")
+		logger.info("seeded editable defaults: %s", path)
+	with open(path, encoding="ascii") as fh:
+		prices = yaml.safe_load(fh)
+	missing = [k for k in HW_PRICE_KEYS if k not in prices]
+	if missing:
+		raise ValueError(f"{path}: missing keys {missing}")
+	return prices
+
+
+def growth(law: str, r: float, n0: int, cores: int) -> float:
+	depth = 1.0 + math.log(r) / math.log(n0)
+	if law == "o1":
+		return 1.0
+	if law == "log":
+		return depth
+	if law == "logk":
+		return r
+	if law == "k_log_n":
+		return r * depth
+	if law == "scan":
+		return r
+	if law == "scan_parallel":
+		return r * BASE_CORES / cores
+	logger.error("growth: unknown law=%r", law)
+	raise ValueError(f"unknown scaling law: {law!r}")
+
+
+def flag_of(seconds: float) -> str:
+	if seconds > FAIL_S:
+		return "FAIL"
+	if seconds > IMPRACTICAL_S:
+		return "IMPRACTICAL"
+	return "OK"
+
+
+def project(fmt: str, qn: int, medians: dict, overhead: dict, r: float, n0: int, cores: int) -> float:
+	variable = max(0.0, medians[(fmt, qn)] - overhead[fmt])
+	return overhead[fmt] + variable * growth(LAWS[(fmt, qn)], r, n0, cores)
+
+
+def pick_pg_cores(medians: dict, overhead: dict, r: float, n0: int) -> int:
+	"""Smallest priced tier; escalate only when it fixes an IMPRACTICAL flag
+	on a parallel-scan cell (the only law that benefits from cores)."""
+	for cores in CORE_OPTIONS:
+		worst = max(project("pg", qn, medians, overhead, r, n0, cores) for qn in QUERY_NUMBERS if LAWS[("pg", qn)] == "scan_parallel")
+		if worst <= IMPRACTICAL_S:
+			return cores
+	return CORE_OPTIONS[-1]
+
+
+def main() -> int:
+	args = parse_task_args("Task 10: extrapolate measurements to 600 GB / 1.2 TB / 6 TB")
+	out_root: Path = args.out_root
+	bench_dir = out_root / "bench"
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		medians = read_medians(bench_dir / "results_median.csv")
+		sizes = read_sizes(bench_dir / "storage_sizes.csv")
+		prices = load_hw_prices(out_root / "config")
+		n0 = expected_counts(load_lab_config(out_root))["friction_cycles"]
+		csv_bytes = sizes[SIZE_VARIANT["csv"]]
+		coeff = {fmt: sizes[SIZE_VARIANT[fmt]] / csv_bytes for fmt in FORMATS}
+		index_share, pg_index_bytes = pg_index_share(get_dsn())
+		logger.info("storage coefficients vs csv: %s; pg index share %.3f",
+			", ".join(f"{fmt}={coeff[fmt]:.2f}" for fmt in FORMATS), index_share)
+		overhead = {fmt: min(medians[(fmt, qn)] for qn in QUERY_NUMBERS) for fmt in FORMATS}
+
+		flags = {"OK": 0, "IMPRACTICAL": 0, "FAIL": 0}
+		with open(bench_dir / "extrapolation.csv", "w", encoding="ascii", newline="") as fh:
+			writer = csv.writer(fh)
+			writer.writerow(["format", "query", "scale_gb", "projected_wall_s", "flag"])
+			for scale_gb in SCALES_GB:
+				r = scale_gb * 1e9 / csv_bytes
+				cores = {fmt: CORE_OPTIONS[0] for fmt in FORMATS}
+				cores["pg"] = pick_pg_cores(medians, overhead, r, n0)
+				for fmt in FORMATS:
+					for qn in QUERY_NUMBERS:
+						seconds = project(fmt, qn, medians, overhead, r, n0, cores[fmt])
+						flag = flag_of(seconds)
+						flags[flag] += 1
+						writer.writerow([fmt, f"q{qn}", scale_gb, round(seconds, 3), flag])
+
+		with open(bench_dir / "hardware_sizing.csv", "w", encoding="ascii", newline="") as fh:
+			writer = csv.writer(fh)
+			writer.writerow(["format", "scale_gb", "disk_tb", "ram_gb", "cores", "nodes",
+				"est_cost_usd", "ram_modules", "nvme_drives", "base_usd", "ram_usd", "disk_usd", "extra_nodes_usd"])
+			for scale_gb in SCALES_GB:
+				r = scale_gb * 1e9 / csv_bytes
+				for fmt in FORMATS:
+					fmt_bytes = coeff[fmt] * scale_gb * 1e9
+					disk_tb = fmt_bytes * DISK_HEADROOM / 1e12
+					if fmt in ("csv", "json"):
+						ram_gb = STREAM_RAM_GB
+					elif fmt in ("sqlite", "pg"):
+						# same index share assumed for sqlite: identical rows,
+						# keys, and indexes as the PG schema (variant B)
+						ram_gb = max(MIN_RAM_GB, CACHE_INDEX_FRACTION * index_share * fmt_bytes / 1e9 + WORKING_SET_GB)
+					else:
+						ram_gb = max(MIN_RAM_GB, RDF_HOT_FRACTION * fmt_bytes / 1e9)
+					cores = pick_pg_cores(medians, overhead, r, n0) if fmt == "pg" else CORE_OPTIONS[0]
+					nodes = max(1, math.ceil(ram_gb / NODE_RAM_CEILING_GB))
+					# bill of materials: whole RDIMM modules and whole NVMe drives
+					ram_modules = math.ceil(ram_gb / prices["ram_module_gb"])
+					nvme_drives = math.ceil(disk_tb / prices["nvme_drive_tb"])
+					base_usd = prices["server_16_core_usd"] if cores == 16 else prices["server_32_core_usd"]
+					ram_usd = ram_modules * prices["ram_module_usd"]
+					disk_usd = nvme_drives * prices["nvme_drive_usd"]
+					extra_nodes_usd = (nodes - 1) * prices["extra_node_usd"]
+					cost = base_usd + ram_usd + disk_usd + extra_nodes_usd
+					writer.writerow([fmt, scale_gb, round(disk_tb, 3), round(ram_gb, 1), cores, nodes,
+						round(cost), ram_modules, nvme_drives,
+						round(base_usd), round(ram_usd), round(disk_usd), round(extra_nodes_usd)])
+
+		entries = {
+			"extrapolation_rows": len(SCALES_GB) * len(FORMATS) * len(QUERY_NUMBERS),
+			"sizing_rows": len(SCALES_GB) * len(FORMATS),
+			"flags_ok": flags["OK"],
+			"flags_impractical": flags["IMPRACTICAL"],
+			"flags_fail": flags["FAIL"],
+			"pg_index_share": round(index_share, 3),
+			"pg_index_mb": round(pg_index_bytes / 1048576, 1),
+			"prices_as_of": prices["as_of"],
+			"note_pandas": "pandas CSV variant not measured; its RAM is O(n) - infeasible above ~10 GB (spec 10)",
+		}
+		for fmt in FORMATS:
+			entries[f"coeff_{fmt}"] = round(coeff[fmt], 3)
+		marker_path = write_marker(out_root, TASK_ID, entries)
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	print(f"task 10 ok: {entries['extrapolation_rows']} projections "
+		f"(ok={flags['OK']}, impractical={flags['IMPRACTICAL']}, fail={flags['FAIL']}), "
+		f"{entries['sizing_rows']} sizing rows")
+	print(f"outputs: {bench_dir / 'extrapolation.csv'}, {bench_dir / 'hardware_sizing.csv'}")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

generate_data.py

@@ -0,0 +1,419 @@
+"""Task 03 - Simulated CSV data generation (~150 MB).
+
+Generates the canonical raw corpus under ./out/csv/ exactly as the
+instruments would write it, driven entirely by out/config/lab_config.yaml
+(volumes, physical-model coefficients, seed). Also writes
+./out/csv/MANIFEST.csv (path, rows, bytes, sha256) and the
+./out/.done/03.ok marker. Spec: docs/specs/03_csv_data_generation.md.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import logging
+import math
+import shutil
+import sys
+from datetime import datetime, timedelta
+from pathlib import Path
+
+import numpy as np
+
+from common.pipeline import (
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	remove_stale_marker,
+	write_marker,
+)
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "03"
+DEPENDS_ON = ["01"]
+
+SIZE_GATE_MIB = (120, 200)  # abort bounds on the generated tree, spec 03
+
+# One float format per column family (docs/rules/data-determinism.md):
+# repr drift between runs must never change file bytes. Precision of the two
+# bulk families also sets corpus size; chosen to land in the 140-170 MB
+# target band of spec 03.
+FMT_COF = "%.16g"
+FMT_POSITION = "%.14g"
+FMT_FORCE = "%.14g"
+FMT_MISC = "%.10g"  # wt%, um, nm, GPa, simtra, wear
+
+# RNG stream tags for per-entity derived generators.
+S_XRF, S_PROF, S_NANO, S_AFM, S_COF, S_LOOP, S_WEAR, S_SIMTRA = range(1, 9)
+
+
+class ValidationError(RuntimeError):
+	pass
+
+
+class CorpusGenerator:
+	def __init__(self, cfg: dict, csv_root: Path) -> None:
+		self.cfg = cfg
+		self.csv_root = csv_root
+		self.seed = cfg["seed"]
+		self.manifest: list[tuple[str, int, int, str]] = []
+		self.total_rows = 0
+		self.total_bytes = 0
+		self.counts = {"coupons": 0, "tracks": 0, "cycle_rows": 0, "loop_rows": 0, "xrf_rows": 0}
+		self.runs_by_batch = {r["batch_code"]: r for r in cfg["runs"]}
+
+	def rng(self, *tags: int) -> np.random.Generator:
+		return np.random.default_rng([self.seed, *tags])
+
+	def write_csv(self, relpath: str, header: str, lines: list[str]) -> None:
+		text = header + "\n" + "\n".join(lines) + "\n"
+		data = text.encode("ascii")
+		path = self.csv_root / relpath
+		path.parent.mkdir(parents=True, exist_ok=True)
+		path.write_bytes(data)
+		sha = hashlib.sha256(data).hexdigest()
+		self.manifest.append((relpath, len(lines), len(data), sha))
+		self.total_rows += len(lines)
+		self.total_bytes += len(data)
+
+	# --- top-level flat files ---
+
+	def gen_batches(self) -> None:
+		sched = self.cfg["schedule"]
+		start = datetime.fromisoformat(sched["deposition_start_date"])
+		lines = []
+		for i, b in enumerate(self.cfg["deposition_matrix"]):
+			date = (start + timedelta(days=i * sched["batch_interval_days"])).date().isoformat()
+			lines.append(
+				f"{b['batch_code']},{b['pt_gun_tilt_deg']},{b['au_gun_tilt_deg']},"
+				f"{b['pt_power_W']},{b['au_power_W']},{b['pt_discharge_V']},{b['au_discharge_V']},{date}"
+			)
+		self.write_csv(
+			"batches.csv",
+			"batch_code,pt_gun_tilt_deg,au_gun_tilt_deg,pt_power_W,au_power_W,pt_discharge_V,au_discharge_V,deposition_date",
+			lines,
+		)
+
+	def gen_runs(self) -> None:
+		plates = self.cfg["friction_assignment"]["plates_per_run"]
+		lines = [
+			f"{r['run_code']},{r['environment']},{r['date']},{plates},{r['operator']}"
+			for r in self.cfg["runs"]
+		]
+		self.write_csv("runs.csv", "run_code,environment,date,plates,operator", lines)
+
+	def gen_simtra(self) -> None:
+		pm = self.cfg["physical_models"]["simtra"]
+		n = self.cfg["volumes"]["simtra_rows_per_batch"]
+		us_pt = pm["surface_binding_energy_eV"]["pt"]
+		us_au = pm["surface_binding_energy_eV"]["au"]
+		for b_idx, b in enumerate(self.cfg["deposition_matrix"], 1):
+			rng = self.rng(b_idx, S_SIMTRA)
+			angle = np.linspace(pm["angle_deg_range"][0], pm["angle_deg_range"][1], n)
+			pt_flux = pm["flux_per_W"] * b["pt_power_W"] * np.cos(np.radians(angle - b["pt_gun_tilt_deg"])) ** 2
+			au_flux = pm["flux_per_W"] * b["au_power_W"] * np.cos(np.radians(angle - b["au_gun_tilt_deg"])) ** 2
+			pt_flux = np.maximum(pt_flux * (1 + rng.normal(0, 0.02, n)), 0)
+			au_flux = np.maximum(au_flux * (1 + rng.normal(0, 0.02, n)), 0)
+			# Thompson mean emission energy ~ 2 * surface binding energy,
+			# flux-weighted across the two species.
+			energy = (pt_flux * 2 * us_pt + au_flux * 2 * us_au) / (pt_flux + au_flux + 1e-12)
+			energy = energy + rng.normal(0, 0.2, n)
+			lines = [
+				f"{FMT_MISC % a},{FMT_MISC % e},{FMT_MISC % p},{FMT_MISC % g}"
+				for a, e, p, g in zip(angle, energy, pt_flux, au_flux)
+			]
+			self.write_csv(f"simtra/simtra_profile_{b['batch_code']}.csv", "angle_deg,energy_eV,pt_flux,au_flux", lines)
+
+	# --- per-coupon measurements ---
+
+	def gen_xrf(self, rel_dir: str, b_idx: int, w: int, c: int, au_nominal: float) -> float:
+		gx_n, gy_n = self.cfg["volumes"]["xrf_grid"]
+		noise_sd = self.cfg["physical_models"]["au_gradient"]["xrf_noise_sd_wtpct"]
+		rng = self.rng(b_idx, w, c, S_XRF)
+		au = au_nominal + rng.normal(0, noise_sd, gx_n * gy_n)
+		lines = []
+		i = 0
+		for gy in range(1, gy_n + 1):
+			for gx in range(1, gx_n + 1):
+				lines.append(f"{gx},{gy},{FMT_MISC % (100.0 - au[i])},{FMT_MISC % au[i]}")
+				i += 1
+		self.write_csv(f"{rel_dir}/xrf_map.csv", "grid_x,grid_y,pt_wtpct,au_wtpct", lines)
+		self.counts["xrf_rows"] += len(lines)
+		return float(au.mean())
+
+	def gen_profilometry(self, rel_dir: str, b_idx: int, w: int, c: int, row: int, col: int) -> float:
+		pm = self.cfg["physical_models"]["thickness_um"]
+		gx_n, gy_n = self.cfg["volumes"]["profilometry_grid"]
+		# Normalized squared wafer radius: corner coupon (r2=1) is thinnest.
+		r2 = (((col - 4) / 3.0) ** 2 + ((row - 4) / 3.0) ** 2) / 2.0
+		nominal = pm["center_um"] - (pm["center_um"] - pm["edge_um"]) * r2
+		rng = self.rng(b_idx, w, c, S_PROF)
+		t = nominal + rng.normal(0, pm["noise_sd_um"], gx_n * gy_n)
+		lines = []
+		i = 0
+		for gy in range(1, gy_n + 1):
+			for gx in range(1, gx_n + 1):
+				lines.append(f"{gx},{gy},{FMT_MISC % t[i]}")
+				i += 1
+		self.write_csv(f"{rel_dir}/profilometry.csv", "grid_x,grid_y,thickness_um", lines)
+		return float(t.mean())
+
+	def gen_nanoindentation(self, rel_dir: str, b_idx: int, w: int, c: int, au_mean: float) -> None:
+		pm = self.cfg["physical_models"]
+		n = self.cfg["volumes"]["nanoindentation_indents_per_coupon"]
+		grid_w = pm["nanoindentation"]["indent_grid"][0]
+		pitch = pm["nanoindentation"]["grid_pitch_um"]
+		max_load = pm["nanoindentation"]["max_load_mN"]
+		rng = self.rng(b_idx, w, c, S_NANO)
+		h = pm["hardness_GPa"]["intercept"] + pm["hardness_GPa"]["slope_per_au_wtpct"] * au_mean \
+			+ rng.normal(0, pm["hardness_GPa"]["noise_sd"], n)
+		er = pm["reduced_modulus_GPa"]["intercept"] + pm["reduced_modulus_GPa"]["slope_per_au_wtpct"] * au_mean \
+			+ rng.normal(0, pm["reduced_modulus_GPa"]["noise_sd"], n)
+		lines = [
+			f"{i + 1},{(i % grid_w) * pitch},{(i // grid_w) * pitch},{FMT_MISC % h[i]},{FMT_MISC % er[i]},{max_load}"
+			for i in range(n)
+		]
+		self.write_csv(
+			f"{rel_dir}/nanoindentation.csv",
+			"indent_id,x_um,y_um,hardness_GPa,reduced_modulus_GPa,max_load_mN",
+			lines,
+		)
+
+	def gen_afm(self, rel_dir: str, b_idx: int, w: int, c: int, coupon_code: str) -> float:
+		pm = self.cfg["physical_models"]["roughness_ra_nm"]
+		rng = self.rng(b_idx, w, c, S_AFM)
+		ra = float(np.exp(rng.normal(math.log(pm["median_nm"]), pm["sigma_log"])))
+		rq = ra * pm["rq_over_ra"]
+		self.write_csv(
+			f"{rel_dir}/afm.csv",
+			"ra_nm,rq_nm,image_file",
+			[f"{FMT_MISC % ra},{FMT_MISC % rq},images/afm/{coupon_code}.tiff"],
+		)
+		return ra
+
+	# --- per-track friction data ---
+
+	def cof_ss(self, au_mean: float, environment: str) -> float:
+		pm = self.cfg["physical_models"]["cof_steady_state"]
+		value = pm[environment]["intercept"] + pm[environment]["slope_per_au_wtpct"] * au_mean
+		return max(value, pm["floor"])  # deliberate model floor, spec 01
+
+	def gen_track(
+		self, rel_dir: str, b_idx: int, w: int, c: int, t: int,
+		track_code: str, run: dict, au_mean: float, plate: int, probe: int, square: int,
+	) -> None:
+		tc = self.cfg["test_conditions"]
+		pm = self.cfg["physical_models"]
+		v = self.cfg["volumes"]
+		sched = self.cfg["schedule"]
+		env = run["environment"]
+		load_mn = tc["normal_load_mN"]
+
+		started = datetime.fromisoformat(f"{run['date']}T{sched['run_start_time']}+00:00") + timedelta(
+			seconds=((plate - 1) * 12 + (square - 1) * 3 + (t - 1)) * sched["track_interval_s"]
+		)
+		counterface_id = f"{run['run_code']}-P{plate}-PR{probe}-B{t}"
+		self.write_csv(
+			f"{rel_dir}/track_info.csv",
+			"track_code,run_code,environment,load_mN,stroke_mm,speed_mm_s,counterface_id,started_at",
+			[
+				f"{track_code},{run['run_code']},{env},{load_mn},{FMT_MISC % tc['stroke_mm']},"
+				f"{FMT_MISC % tc['speed_mm_s']},{counterface_id},{started.strftime('%Y-%m-%dT%H:%M:%SZ')}"
+			],
+		)
+
+		cof_ss = self.cof_ss(au_mean, env)
+		cof_rng = self.rng(b_idx, w, c, t, S_COF)
+		ri = pm["cof_run_in"]
+		cof_0 = cof_rng.uniform(*ri["cof_0_range"])
+		tau = cof_rng.uniform(*ri["tau_cycles_range"])
+		n_cycles = v["cycles_per_track"]
+		cycles = np.arange(1, n_cycles + 1)
+		decay = (cof_0 - cof_ss) * np.exp(-cycles / tau)
+		cof = cof_ss + decay + cof_rng.normal(0, ri["noise_sd"], n_cycles)
+		self.write_csv(
+			f"{rel_dir}/cof_vs_cycle.csv",
+			"cycle,cof",
+			[f"{cyc},{FMT_COF % val}" for cyc, val in zip(cycles, cof)],
+		)
+		self.counts["cycle_rows"] += n_cycles
+
+		loop_rng = self.rng(b_idx, w, c, t, S_LOOP)
+		half_um = tc["stroke_mm"] * 1000.0 / 2.0
+		pts_half = v["loop_points_per_loop"] // 2
+		fwd = np.linspace(-half_um, half_um, pts_half)
+		rev = np.linspace(half_um, -half_um, pts_half)
+		force_sd = pm["friction_loop"]["force_noise_sd_mN"]
+		lines: list[str] = []
+		for k in range(1, v["loops_per_track"] + 1):
+			cyc = k * v["loop_every_n_cycles"]
+			mu = cof_ss + (cof_0 - cof_ss) * math.exp(-cyc / tau)
+			f_fwd = mu * load_mn + loop_rng.normal(0, force_sd, pts_half)
+			f_rev = -mu * load_mn + loop_rng.normal(0, force_sd, pts_half)
+			lines.extend(f"{cyc},{FMT_POSITION % p},{FMT_FORCE % f}" for p, f in zip(fwd, f_fwd))
+			lines.extend(f"{cyc},{FMT_POSITION % p},{FMT_FORCE % f}" for p, f in zip(rev, f_rev))
+		self.write_csv(f"{rel_dir}/friction_loops.csv", "cycle,position_um,friction_force_mN", lines)
+		self.counts["loop_rows"] += len(lines)
+
+		wm = pm["wear_archard"]
+		wear_rng = self.rng(b_idx, w, c, t, S_WEAR)
+		k_arch = wm["k0_mm3_per_N_m"][env] * (1 + wm["slope_per_au_wtpct"][env] * au_mean)
+		k_arch *= math.exp(wear_rng.normal(0, wm["noise_sigma_log"]))
+		k_arch = max(k_arch, wm["k_floor_mm3_per_N_m"])  # deliberate model floor, spec 01
+		sliding_m = n_cycles * 2 * tc["stroke_mm"] / 1000.0
+		volume_um3 = k_arch * (load_mn / 1000.0) * sliding_m * 1e9
+		self.write_csv(
+			f"{rel_dir}/wear.csv",
+			"wear_volume_um3,k_archard,sliding_distance_m",
+			[f"{FMT_MISC % volume_um3},{FMT_MISC % k_arch},{FMT_MISC % sliding_m}"],
+		)
+		self.counts["tracks"] += 1
+
+	# --- hierarchy walk ---
+
+	def generate(self) -> None:
+		self.gen_batches()
+		self.gen_runs()
+		self.gen_simtra()
+
+		h = self.cfg["hierarchy"]
+		fa = self.cfg["friction_assignment"]
+		ag = self.cfg["physical_models"]["au_gradient"]
+		sched = self.cfg["schedule"]
+		dep_start = datetime.fromisoformat(sched["deposition_start_date"])
+		reserve = set(fa["reserve_grid_positions"])
+		grid_w = h["coupon_grid_per_wafer"][0]
+
+		for b_idx, batch in enumerate(self.cfg["deposition_matrix"], 1):
+			b_code = batch["batch_code"]
+			run = self.runs_by_batch[b_code]
+			au_center = ag["batch_center_wtpct"][b_code]
+			dep_date = (dep_start + timedelta(days=(b_idx - 1) * sched["batch_interval_days"])).date().isoformat()
+			friction_idx = 0
+			for w in range(1, h["wafers_per_batch"] + 1):
+				wafer_code = f"{b_code}-W{w}"
+				wafer_dir = f"batch_{b_code}/wafer_W{w}"
+				self.write_csv(
+					f"{wafer_dir}/wafer_info.csv",
+					"wafer_code,batch_code,wafer_index,deposition_date,coupons,friction_coupons,reserve_coupons",
+					[
+						f"{wafer_code},{b_code},{w},{dep_date},{h['coupons_per_wafer']},"
+						f"{h['coupons_per_wafer'] - len(reserve)},{len(reserve)}"
+					],
+				)
+				for c in range(1, h["coupons_per_wafer"] + 1):
+					row = (c - 1) // grid_w + 1
+					col = (c - 1) % grid_w + 1
+					coupon_code = f"{wafer_code}-C{c:02d}"
+					rel_dir = f"{wafer_dir}/coupon_C{c:02d}"
+					au_nominal = au_center + ag["wafer_span_wtpct"] * (col - 4) / 3.0
+
+					au_mean = self.gen_xrf(rel_dir, b_idx, w, c, au_nominal)
+					thickness = self.gen_profilometry(rel_dir, b_idx, w, c, row, col)
+					self.gen_nanoindentation(rel_dir, b_idx, w, c, au_mean)
+					ra = self.gen_afm(rel_dir, b_idx, w, c, coupon_code)
+
+					if c in reserve:
+						assignment = "RESERVE,,,"
+					else:
+						plate = friction_idx // 24 + 1
+						probe = (friction_idx % 24) // 4 + 1
+						square = friction_idx % 4 + 1
+						friction_idx += 1
+						assignment = f"{run['run_code']},{plate},{probe},{square}"
+					self.write_csv(
+						f"{rel_dir}/coupon_info.csv",
+						"coupon_code,batch_code,wafer_code,grid_row,grid_col,thickness_um,ra_nm,au_wtpct_mean,run_code,plate,probe,square",
+						[
+							f"{coupon_code},{b_code},{wafer_code},{row},{col},{FMT_MISC % thickness},"
+							f"{FMT_MISC % ra},{FMT_MISC % au_mean},{assignment}"
+						],
+					)
+					self.counts["coupons"] += 1
+
+					if c not in reserve:
+						plate, probe, square = (int(x) for x in assignment.split(",")[1:])
+						for t in range(1, fa["tracks_per_friction_coupon"] + 1):
+							self.gen_track(
+								f"{rel_dir}/track_T{t}", b_idx, w, c, t,
+								f"{coupon_code}-T{t}", run, au_mean, plate, probe, square,
+							)
+				logger.debug("wafer %s done", wafer_code)
+			logger.info(
+				"batch %s done: %d files, %.1f MiB cumulative",
+				b_code, len(self.manifest), self.total_bytes / (1024 * 1024),
+			)
+
+	def validate_counts(self) -> None:
+		v = self.cfg["volumes"]
+		expected = {
+			"coupons": v["coupons_total"],
+			"tracks": v["tracks_total"],
+			"cycle_rows": v["cycle_rows_total"],
+			"loop_rows": v["loop_points_total"],
+			"xrf_rows": v["xrf_points_total"],
+		}
+		for key, exp in expected.items():
+			if self.counts[key] != exp:
+				raise ValidationError(f"count mismatch: {key}: generated {self.counts[key]} != expected {exp}")
+		logger.info("all entity counts match lab_config volumes: %s", self.counts)
+
+	def write_manifest(self) -> int:
+		lines = [f"{p},{r},{b},{s}" for p, r, b, s in self.manifest]
+		text = "path,rows,bytes,sha256\n" + "\n".join(lines) + "\n"
+		data = text.encode("ascii")
+		(self.csv_root / "MANIFEST.csv").write_bytes(data)
+		return len(data)
+
+
+def main() -> int:
+	args = parse_task_args("Task 03: generate the CSV corpus")
+	out_root: Path = args.out_root
+	csv_root = out_root / "csv"
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		if csv_root.exists():
+			logger.info("re-run: removing previous corpus %s", csv_root)
+			shutil.rmtree(csv_root)
+		cfg = load_lab_config(out_root)
+
+		gen = CorpusGenerator(cfg, csv_root)
+		gen.generate()
+		gen.validate_counts()
+		manifest_bytes = gen.write_manifest()
+
+		tree_bytes = gen.total_bytes + manifest_bytes
+		tree_mib = tree_bytes / (1024 * 1024)
+		print(f"generated tree size: {tree_mib:.1f} MiB ({tree_bytes} bytes, {len(gen.manifest) + 1} files)")
+		if not SIZE_GATE_MIB[0] <= tree_mib <= SIZE_GATE_MIB[1]:
+			raise ValidationError(
+				f"tree size {tree_mib:.1f} MiB outside gate {SIZE_GATE_MIB[0]}-{SIZE_GATE_MIB[1]} MiB"
+			)
+		marker_path = write_marker(out_root, TASK_ID, {
+			"csv_root": gen.csv_root.as_posix(),
+			"files": len(gen.manifest),
+			"total_rows": gen.total_rows,
+			"total_bytes": tree_bytes,
+			"coupons": gen.counts["coupons"],
+			"tracks": gen.counts["tracks"],
+			"cycle_rows": gen.counts["cycle_rows"],
+			"loop_points": gen.counts["loop_rows"],
+			"xrf_points": gen.counts["xrf_rows"],
+		})
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	c = gen.counts
+	print(
+		f"rows: total={gen.total_rows} cycles={c['cycle_rows']} loops={c['loop_rows']} "
+		f"xrf={c['xrf_rows']} coupons={c['coupons']} tracks={c['tracks']}"
+	)
+	print(f"manifest: {csv_root / 'MANIFEST.csv'} ({len(gen.manifest)} entries)")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

make_lab_config.py

@@ -0,0 +1,343 @@
+"""Task 01 - Laboratory configuration.
+
+Writes ./out/config/lab_config.yaml, the authoritative machine-readable
+laboratory model (volumes, deposition matrix, physical-model coefficients,
+random seed) that all downstream tasks read, plus the ./out/.done/01.ok
+completion marker. Spec: docs/specs/01_lab_configuration.md.
+"""
+
+from __future__ import annotations
+
+import logging
+import sys
+from pathlib import Path
+
+import yaml
+
+from common.pipeline import parse_task_args, remove_stale_marker, write_marker
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "01"
+
+# Transcription of docs/specs/01_lab_configuration.md. This literal is the
+# ONLY sanctioned home of these values in code; every other task reads the
+# YAML this script writes (docs/rules/code-config-yaml.md).
+LAB_CONFIG: dict = {
+	"project": "LabDataStorageEvaluation",
+	"description": "Simulated tribology laboratory (Sandia Pt-Au LDRD context)",
+	"spec": "docs/specs/01_lab_configuration.md",
+	"seed": 20260711,
+	"material_system": {
+		"substrate": {"material": "Ti-6Al-4V", "dimensions_mm": [10, 10, 3]},
+		"adhesion_layer": {"material": "Cr", "process": "sputtered"},
+		"coating": {
+			"material": "Pt-Au",
+			"deposition": "composition gradient across each wafer",
+			"thickness_um_range": [0.3, 1.1],
+		},
+	},
+	"instruments": [
+		{
+			"instrument_id": "rapid",
+			"role": "friction",
+			"name": "RAPID custom high-throughput parallelized 6-probe tribometer",
+			"data_produced": "COF vs cycle per track; test conditions",
+		},
+		{
+			"instrument_id": "ti980",
+			"role": "nanoindentation",
+			"name": "Bruker TI980 TriboIndenter",
+			"data_produced": "hardness, reduced modulus (25 indents per coupon)",
+		},
+		{
+			"instrument_id": "m4_tornado",
+			"role": "composition",
+			"name": "Bruker M4 Tornado micro-XRF",
+			"data_produced": "Pt/Au wt% map per coupon (20x20 grid)",
+		},
+		{
+			"instrument_id": "pvd200",
+			"role": "deposition",
+			"name": "Kurt J. Lesker PVD 200 sputter-down",
+			"data_produced": "batch deposition parameters",
+		},
+		{
+			"instrument_id": "afm",
+			"role": "surface_roughness",
+			"name": "AFM",
+			"data_produced": "topography image metadata + Ra/Rq per coupon",
+		},
+		{
+			"instrument_id": "profilometer",
+			"role": "film_thickness",
+			"name": "Optical profilometry",
+			"data_produced": "thickness map per coupon (10x10 grid)",
+		},
+		{
+			"instrument_id": "simtra",
+			"role": "simulation",
+			"name": "SIMTRA sputter transport Monte-Carlo",
+			"data_produced": "deposition atom-energy / composition profiles per deposition run",
+		},
+	],
+	"deposition_matrix": [
+		{
+			"batch_code": "B721",
+			"pt_gun_tilt_deg": 20,
+			"au_gun_tilt_deg": 0,
+			"pt_power_W": 150,
+			"au_power_W": 50,
+			"pt_discharge_V": 432,
+			"au_discharge_V": 311,
+		},
+		{
+			"batch_code": "B722",
+			"pt_gun_tilt_deg": 20,
+			"au_gun_tilt_deg": 20,
+			"pt_power_W": 100,
+			"au_power_W": 100,
+			"pt_discharge_V": 399,
+			"au_discharge_V": 352,
+		},
+		{
+			"batch_code": "B723",
+			"pt_gun_tilt_deg": 20,
+			"au_gun_tilt_deg": 20,
+			"pt_power_W": 150,
+			"au_power_W": 50,
+			"pt_discharge_V": 430,
+			"au_discharge_V": 311,
+		},
+		{
+			"batch_code": "B724",
+			"pt_gun_tilt_deg": 0,
+			"au_gun_tilt_deg": 20,
+			"pt_power_W": 50,
+			"au_power_W": 150,
+			"pt_discharge_V": 376,
+			"au_discharge_V": 340,
+		},
+	],
+	"hierarchy": {
+		"batches": 4,
+		"wafers_per_batch": 3,
+		"coupon_grid_per_wafer": [7, 7],
+		"coupons_per_wafer": 49,
+	},
+	"friction_assignment": {
+		"friction_coupons_total": 480,
+		"reserve_coupons_total": 108,
+		"runs": 4,
+		"coupons_per_run": 120,
+		"plates_per_run": 5,
+		"probes_per_plate": 6,
+		"coupons_per_probe_square": 4,
+		"tracks_per_friction_coupon": 3,
+		"counterfaces_per_holder": 3,
+		"fresh_counterface_per_track": True,
+		# QA witness pattern: rows 1/4/7 x cols 1/4/7 of the 7x7 grid stay
+		# unworn (reserve), sampling low/mid/high Au columns on every wafer.
+		"reserve_grid_positions": [1, 4, 7, 22, 25, 28, 43, 46, 49],
+	},
+	"volumes": {
+		"coupons_total": 588,
+		"tracks_total": 1440,
+		"cycles_per_track": 1000,
+		"cycle_rows_total": 1440000,
+		"loop_every_n_cycles": 100,
+		"loops_per_track": 10,
+		"loop_points_per_loop": 200,
+		"loop_points_total": 2880000,
+		"xrf_grid": [20, 20],
+		"xrf_points_per_coupon": 400,
+		"xrf_points_total": 235200,
+		"profilometry_grid": [10, 10],
+		"profilometry_points_per_coupon": 100,
+		"nanoindentation_indents_per_coupon": 25,
+		"simtra_rows_per_batch": 1000,
+	},
+	"runs": [
+		{"run_code": "R1", "batch_code": "B721", "environment": "lab_air", "date": "2026-04-06", "operator": "A. Reyes"},
+		{"run_code": "R2", "batch_code": "B722", "environment": "lab_air", "date": "2026-04-13", "operator": "K. Patel"},
+		{"run_code": "R3", "batch_code": "B723", "environment": "dry_n2", "date": "2026-04-20", "operator": "A. Reyes"},
+		{"run_code": "R4", "batch_code": "B724", "environment": "dry_n2", "date": "2026-04-27", "operator": "M. Novak"},
+	],
+	"schedule": {
+		"deposition_start_date": "2026-03-02",
+		"batch_interval_days": 7,
+		"characterization_lag_days": 3,
+		"run_start_time": "09:00:00",
+		"track_interval_s": 2100,
+	},
+	"test_conditions": {
+		"normal_load_mN": 100,
+		"stroke_mm": 1.0,
+		"speed_mm_s": 1.0,
+		"counterface": {"material": "ruby (Al2O3)", "diameter_mm": 3.175},
+		"rh_pct": {"lab_air": 45, "dry_n2": 2},
+		"temperature_C": 23,
+		"temperature_tolerance_C": 1,
+	},
+	"physical_models": {
+		"au_gradient": {
+			"description": "linear Au wt% gradient along wafer x, batch-dependent center",
+			"batch_center_wtpct": {"B721": 8, "B722": 25, "B723": 10, "B724": 60},
+			"wafer_span_wtpct": 5,
+			"xrf_noise_sd_wtpct": 0.3,
+		},
+		"cof_run_in": {
+			"formula": "cof(c) = cof_ss + (cof_0 - cof_ss) * exp(-c / tau) + N(0, sigma)",
+			"cof_0_range": [0.35, 0.5],
+			"tau_cycles_range": [100, 400],
+			"noise_sd": 0.01,
+		},
+		"cof_steady_state": {
+			"formula": "cof_ss = intercept + slope_per_au_wtpct * au_wtpct_mean, clamped to >= floor",
+			"lab_air": {"intercept": 0.32, "slope_per_au_wtpct": -0.0015},
+			"dry_n2": {"intercept": 0.24, "slope_per_au_wtpct": -0.0012},
+			"floor": 0.12,
+		},
+		"hardness_GPa": {
+			"formula": "H = intercept + slope_per_au_wtpct * au_wtpct_mean + N(0, noise_sd)",
+			"intercept": 8.5,
+			"slope_per_au_wtpct": -0.05,
+			"noise_sd": 0.25,
+		},
+		"reduced_modulus_GPa": {
+			"formula": "Er = intercept + slope_per_au_wtpct * au_wtpct_mean + N(0, noise_sd)",
+			"intercept": 190,
+			"slope_per_au_wtpct": -0.6,
+			"noise_sd": 4,
+		},
+		"roughness_ra_nm": {
+			"distribution": "lognormal",
+			"median_nm": 5,
+			"sigma_log": 0.3,
+			"rq_over_ra": 1.25,
+		},
+		"thickness_um": {
+			"profile": "radial parabolic across wafer plus noise",
+			"center_um": 1.05,
+			"edge_um": 0.35,
+			"noise_sd_um": 0.02,
+			"range_um": [0.3, 1.1],
+		},
+		"wear_archard": {
+			# Spec 01 fixes only the model shape (V = k*F*s, k depends on Au%
+			# and environment); these coefficients are project defaults chosen
+			# to yield Sandia-plausible low-wear volumes at 100 mN / 2 m.
+			"formula": "V = k * F * s; k = k0 * (1 + slope_per_au_wtpct * au_wtpct_mean), lognormal noise, clamped to >= k_floor",
+			"k0_mm3_per_N_m": {"lab_air": 1.5e-07, "dry_n2": 6.0e-08},
+			"slope_per_au_wtpct": {"lab_air": -0.008, "dry_n2": -0.010},
+			"k_floor_mm3_per_N_m": 1.0e-09,
+			"noise_sigma_log": 0.15,
+		},
+		"friction_loop": {
+			"force_noise_sd_mN": 0.3,
+		},
+		"nanoindentation": {
+			"max_load_mN": 10,
+			"indent_grid": [5, 5],
+			"grid_pitch_um": 20,
+		},
+		"simtra": {
+			"angle_deg_range": [0, 90],
+			"energy_eV_max": 50,
+			"surface_binding_energy_eV": {"pt": 5.84, "au": 3.81},
+			"flux_per_W": 0.02,
+		},
+	},
+}
+
+YAML_HEADER = (
+	"# Generated by make_lab_config.py (task 01) from docs/specs/01_lab_configuration.md.\n"
+	"# Do not hand-edit: regenerate via `python make_lab_config.py`.\n"
+)
+
+
+def validate_config(cfg: dict) -> None:
+	"""Assert the arithmetic identities between declared volumes (spec 01)."""
+	h = cfg["hierarchy"]
+	f = cfg["friction_assignment"]
+	v = cfg["volumes"]
+	checks = [
+		("coupons_total", v["coupons_total"], h["batches"] * h["wafers_per_batch"] * h["coupons_per_wafer"]),
+		("coupons_per_wafer", h["coupons_per_wafer"], h["coupon_grid_per_wafer"][0] * h["coupon_grid_per_wafer"][1]),
+		("friction+reserve", v["coupons_total"], f["friction_coupons_total"] + f["reserve_coupons_total"]),
+		("friction_coupons_total", f["friction_coupons_total"], f["runs"] * f["coupons_per_run"]),
+		("coupons_per_run", f["coupons_per_run"], f["plates_per_run"] * f["probes_per_plate"] * f["coupons_per_probe_square"]),
+		("tracks_total", v["tracks_total"], f["friction_coupons_total"] * f["tracks_per_friction_coupon"]),
+		("cycle_rows_total", v["cycle_rows_total"], v["tracks_total"] * v["cycles_per_track"]),
+		("loops_per_track", v["loops_per_track"], v["cycles_per_track"] // v["loop_every_n_cycles"]),
+		("loop_points_total", v["loop_points_total"], v["tracks_total"] * v["loops_per_track"] * v["loop_points_per_loop"]),
+		("xrf_points_per_coupon", v["xrf_points_per_coupon"], v["xrf_grid"][0] * v["xrf_grid"][1]),
+		("xrf_points_total", v["xrf_points_total"], v["coupons_total"] * v["xrf_points_per_coupon"]),
+		("profilometry_points_per_coupon", v["profilometry_points_per_coupon"], v["profilometry_grid"][0] * v["profilometry_grid"][1]),
+		("runs_count", f["runs"], len(cfg["runs"])),
+		("batches_count", h["batches"], len(cfg["deposition_matrix"])),
+		("reserve_total", f["reserve_coupons_total"], len(f["reserve_grid_positions"]) * h["wafers_per_batch"] * h["batches"]),
+		("friction_per_batch", f["coupons_per_run"], (h["coupons_per_wafer"] - len(f["reserve_grid_positions"])) * h["wafers_per_batch"]),
+	]
+	for name, declared, derived in checks:
+		if declared != derived:
+			raise ValueError(f"volume identity failed: {name}: declared {declared} != derived {derived}")
+	run_batches = [r["batch_code"] for r in cfg["runs"]]
+	matrix_batches = [b["batch_code"] for b in cfg["deposition_matrix"]]
+	if sorted(run_batches) != sorted(matrix_batches):
+		raise ValueError(f"run-batch mapping mismatch: runs cover {run_batches}, matrix has {matrix_batches}")
+	logger.info("all %d volume identities hold", len(checks))
+
+
+def write_yaml(cfg: dict, config_path: Path) -> int:
+	config_path.parent.mkdir(parents=True, exist_ok=True)
+	body = yaml.safe_dump(cfg, sort_keys=False, allow_unicode=False, width=100)
+	text = YAML_HEADER + body
+	with open(config_path, "w", encoding="utf-8", newline="\n") as fh:
+		fh.write(text)
+	reloaded = yaml.safe_load(text)
+	if reloaded != cfg:
+		raise RuntimeError(f"YAML round-trip mismatch for {config_path}")
+	return len(text.encode("utf-8"))
+
+
+def main() -> int:
+	args = parse_task_args("Task 01: write out/config/lab_config.yaml")
+	out_root: Path = args.out_root
+	config_path = out_root / "config" / "lab_config.yaml"
+	try:
+		remove_stale_marker(out_root, TASK_ID)
+		validate_config(LAB_CONFIG)
+		config_bytes = write_yaml(LAB_CONFIG, config_path)
+		v = LAB_CONFIG["volumes"]
+		f = LAB_CONFIG["friction_assignment"]
+		h = LAB_CONFIG["hierarchy"]
+		marker_path = write_marker(out_root, TASK_ID, {
+			"config_file": config_path.as_posix(),
+			"config_bytes": config_bytes,
+			"seed": LAB_CONFIG["seed"],
+			"batches": h["batches"],
+			"wafers": h["batches"] * h["wafers_per_batch"],
+			"coupons": v["coupons_total"],
+			"friction_coupons": f["friction_coupons_total"],
+			"tracks": v["tracks_total"],
+			"cycle_rows": v["cycle_rows_total"],
+			"loop_points": v["loop_points_total"],
+			"xrf_points": v["xrf_points_total"],
+		})
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	v = LAB_CONFIG["volumes"]
+	print(f"task 01 ok: {config_path} ({config_bytes} bytes)")
+	print(
+		f"volumes: batches=4 wafers=12 coupons={v['coupons_total']} tracks={v['tracks_total']} "
+		f"cycle_rows={v['cycle_rows_total']} loop_points={v['loop_points_total']} xrf_points={v['xrf_points_total']}"
+	)
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

make_process_flow.py

@@ -0,0 +1,260 @@
+"""Task 02 - Process flow diagrams.
+
+Reads ./out/config/lab_config.yaml and writes the Mermaid documentation of
+the laboratory workflow: ./out/report/process_flow.md plus one standalone
+./out/report/diagrams/D<n>.mermaid file per diagram (D1-D6), and the
+./out/.done/02.ok completion marker. Spec: docs/specs/02_process_flow_diagrams.md.
+"""
+
+from __future__ import annotations
+
+import logging
+import sys
+from pathlib import Path
+
+from common.pipeline import (
+	check_dependencies,
+	load_lab_config,
+	parse_task_args,
+	remove_stale_marker,
+	write_marker,
+)
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "02"
+DEPENDS_ON = ["01"]
+
+DIAGRAM_TYPES = ("flowchart", "sequenceDiagram", "erDiagram")
+
+
+def d1_coupon_assembly(cfg: dict) -> str:
+	sub = cfg["material_system"]["substrate"]
+	dims = "x".join(str(d) for d in sub["dimensions_mm"])
+	return f"""flowchart LR
+    A([{sub['material']} Base {dims} mm]) --> B[Cleaning]
+    B --> B1([Rinsed in deionized water])
+    B1 --> B2([Sonicated in cleaning solution])
+    B --> C[Smearing Adhesive]
+    C --> C1([{cfg['material_system']['adhesion_layer']['material']} adhesive coating])
+    C --> D[Vapor Deposition System<br/>Kurt J. Lesker PVD 200]
+    D --> D1([PtAu sputter coating, gradient])
+    D --> E([Test Coupon])
+"""
+
+
+def d2_characterization_batch_assembly(cfg: dict) -> str:
+	h = cfg["hierarchy"]
+	t_lo, t_hi = cfg["material_system"]["coating"]["thickness_um_range"]
+	lines = [
+		"flowchart LR",
+		"    TC([Test Coupon]) --> S[SIMTRA simulation<br/>composition / atom energies]",
+		f"    S --> OP[Optical profilometry<br/>film thickness {t_lo}-{t_hi} um]",
+		f"    OP --> TW[Assemble Test Wafer<br/>x{h['coupons_per_wafer']} coupons]",
+		f"    TW --> TB[Assemble Test Batch<br/>x{h['wafers_per_batch']} wafers]",
+	]
+	for row in cfg["deposition_matrix"]:
+		code = row["batch_code"]
+		lines.append(
+			f"    TB --> {code}[{code}<br/>Pt {row['pt_power_W']} W / Au {row['au_power_W']} W]"
+		)
+	return "\n".join(lines) + "\n"
+
+
+def d3_testing_tree(cfg: dict) -> str:
+	v = cfg["volumes"]
+	xrf = "x".join(str(g) for g in v["xrf_grid"])
+	return f"""flowchart TD
+    TC([Test Coupon]) --> F[Friction<br/>RAPID 6-probe tribometer]
+    F --> FA([Lab Air -> COF dataset mu_normal])
+    F --> FN([Dry N2 -> COF dataset mu_dry_nit])
+    TC --> NI[Nanoindentation<br/>Bruker TI980]
+    NI --> NIR([Hardness, reduced modulus<br/>{v['nanoindentation_indents_per_coupon']} indents per coupon])
+    TC --> AF[AFM]
+    AF --> AFR([Topography -> Ra / Rq])
+    TC --> XR[micro-XRF<br/>Bruker M4 Tornado]
+    XR --> XRR([Pt/Au wt% map, {xrf} grid])
+"""
+
+
+def d4_tribometer_session(cfg: dict) -> str:
+	f = cfg["friction_assignment"]
+	return f"""sequenceDiagram
+    actor Op as Operator
+    participant Prep as Sample Prep
+    participant Plan as Excel Test Plan
+    participant SW as Control Software
+    participant TR as RAPID Tribometer
+
+    Op->>Prep: Request test
+    Op->>Prep: Get samples
+    Op->>Prep: Load samples onto plates
+    Op->>Prep: Load ball holders ({f['counterfaces_per_holder']} counterfaces per holder)
+    Op->>Plan: Create test plan<br/>(Sample Plate, Plate Location, Sample ID, Save Location,<br/>Folder Name, X/Y/Z Offset, Load, Iterations)
+    Op->>TR: Transfer plates to tribometer
+    Op->>TR: Set up platter
+    Op->>SW: Activate software
+    Op->>SW: Load plan
+    SW->>TR: Activate tribometer
+    TR-->>SW: Run reciprocating tests (execution loop, see D5)
+    Op->>SW: Software stop
+    SW-->>Op: Save avg files
+    Op->>TR: Remove plates and platters
+    Op->>SW: Close software
+    Op->>TR: Pull equipment out
+"""
+
+
+def d5_execution_loop(cfg: dict) -> str:
+	f = cfg["friction_assignment"]
+	tracks_per_run = f["coupons_per_run"] * f["tracks_per_friction_coupon"]
+	return f"""flowchart TD
+    L[Load all counterfaces] --> P{{{f['plates_per_run']} plates}}
+    P -->|for each plate| PR{{{f['probes_per_plate']} probes in parallel}}
+    PR -->|for each probe| C{{{f['coupons_per_probe_square']} coupons on coupon square}}
+    C -->|for each coupon| T[Draw track]
+    T --> RC[Rotate counterface - fresh ball per track]
+    RC -->|{f['tracks_per_friction_coupon']} tracks per coupon| T
+    T --> DONE([{f['coupons_per_run']} coupons x {f['tracks_per_friction_coupon']} tracks = {tracks_per_run} tracks per run])
+"""
+
+
+def d6_data_hierarchy(cfg: dict) -> str:
+	return """erDiagram
+    DEPOSITION_RUN ||--|| SIMTRA_PROFILE : produces
+    DEPOSITION_RUN ||--|| BATCH : creates
+    BATCH ||--|{ WAFER : contains
+    WAFER ||--|{ COUPON : contains
+    COUPON ||--|| XRF_MAP : has
+    COUPON ||--|| NANOINDENTATION : has
+    COUPON ||--|| AFM : has
+    COUPON ||--|| PROFILOMETRY : has
+    COUPON ||--o{ TRACK : "friction coupons only"
+    TRACK ||--|{ CYCLE : has
+    TRACK ||--|{ LOOP_POINT : has
+    TRACK ||--|| WEAR : has
+"""
+
+
+def deposition_table(cfg: dict) -> str:
+	lines = [
+		"| Batch | Pt:Au gun tilt | Pt power W | Au power W | Pt discharge V | Au discharge V |",
+		"|---|---|---|---|---|---|",
+	]
+	for r in cfg["deposition_matrix"]:
+		lines.append(
+			f"| {r['batch_code']} | {r['pt_gun_tilt_deg']}:{r['au_gun_tilt_deg']} deg "
+			f"| {r['pt_power_W']} | {r['au_power_W']} | {r['pt_discharge_V']} | {r['au_discharge_V']} |"
+		)
+	return "\n".join(lines) + "\n"
+
+
+def build_diagrams(cfg: dict) -> list[tuple[str, str, str, str]]:
+	"""Return (file_stem, title, description, mermaid_text) per diagram."""
+	h = cfg["hierarchy"]
+	return [
+		(
+			"D1",
+			"Coupon Assembly",
+			"Substrate preparation, adhesion layer, and gradient Pt-Au sputter deposition.",
+			d1_coupon_assembly(cfg),
+		),
+		(
+			"D2",
+			"Characterization and Batch Assembly",
+			f"Per-coupon simulation and thickness characterization, then assembly into wafers "
+			f"(x{h['coupons_per_wafer']} coupons) and batches (x{h['wafers_per_batch']} wafers), "
+			f"deposited per the parameter table below.",
+			d2_characterization_batch_assembly(cfg),
+		),
+		(
+			"D3",
+			"Testing Tree",
+			"The four characterization/testing paths every coupon can take.",
+			d3_testing_tree(cfg),
+		),
+		(
+			"D4",
+			"Tribometer Session Sequence",
+			"Operator workflow for one RAPID tribometer session, from test request to teardown.",
+			d4_tribometer_session(cfg),
+		),
+		(
+			"D5",
+			"Execution Loop",
+			"Nested iteration executed by the tribometer within one run.",
+			d5_execution_loop(cfg),
+		),
+		(
+			"D6",
+			"Data Hierarchy",
+			"Entity containment and per-entity measurement datasets (ERD-style).",
+			d6_data_hierarchy(cfg),
+		),
+	]
+
+
+def validate_diagram(stem: str, text: str) -> None:
+	first = text.splitlines()[0].strip() if text.strip() else ""
+	if not first.startswith(DIAGRAM_TYPES):
+		raise ValueError(f"{stem}: unexpected diagram header {first!r}")
+
+
+def write_outputs(cfg: dict, out_root: Path) -> tuple[Path, int, int]:
+	diagrams_dir = out_root / "report" / "diagrams"
+	diagrams_dir.mkdir(parents=True, exist_ok=True)
+	diagrams = build_diagrams(cfg)
+
+	md_parts = [
+		"# Process Flow - Simulated Tribology Laboratory\n",
+		"Generated by make_process_flow.py (task 02) from out/config/lab_config.yaml.",
+		"Standalone diagram sources: ./diagrams/D1..D6.mermaid.\n",
+	]
+	total_diagram_bytes = 0
+	for stem, title, description, text in diagrams:
+		validate_diagram(stem, text)
+		path = diagrams_dir / f"{stem}.mermaid"
+		with open(path, "w", encoding="utf-8", newline="\n") as fh:
+			fh.write(text)
+		total_diagram_bytes += len(text.encode("utf-8"))
+		logger.info("wrote %s (%s)", path, title)
+		md_parts.append(f"## {stem} - {title}\n")
+		md_parts.append(description + "\n")
+		md_parts.append(f"```mermaid\n{text}```\n")
+		if stem == "D2":
+			md_parts.append("Deposition matrix:\n")
+			md_parts.append(deposition_table(cfg))
+
+	md_text = "\n".join(md_parts)
+	md_path = out_root / "report" / "process_flow.md"
+	with open(md_path, "w", encoding="utf-8", newline="\n") as fh:
+		fh.write(md_text)
+	return md_path, len(md_text.encode("utf-8")), total_diagram_bytes
+
+
+def main() -> int:
+	args = parse_task_args("Task 02: write process flow diagrams")
+	out_root: Path = args.out_root
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		cfg = load_lab_config(out_root)
+		md_path, md_bytes, diagram_bytes = write_outputs(cfg, out_root)
+		marker_path = write_marker(out_root, TASK_ID, {
+			"process_flow_file": md_path.as_posix(),
+			"process_flow_bytes": md_bytes,
+			"diagrams": 6,
+			"diagram_bytes": diagram_bytes,
+		})
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	print(f"task 02 ok: {md_path} ({md_bytes} bytes)")
+	print(f"diagrams: 6 files under {out_root / 'report' / 'diagrams'} ({diagram_bytes} bytes)")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

make_queries.py

@@ -0,0 +1,339 @@
+"""Task 08 - Benchmark query definitions (Q1-Q7 x 5 formats).
+
+Writes the runnable query files under ./out/bench/queries/ (SQL for
+SQLite/PostgreSQL, SPARQL text + python drivers for RDF, python drivers
+for CSV/JSON-LD), then produces the canonical expected results: executed
+on PostgreSQL, cross-validated against SQLite row-by-row within 1e-9, and
+stored as ./out/bench/expected/q<N>.rows.csv + q<N>.sha256. CSV, JSON and
+RDF implementations are spot-checked on Q1. Writes ./out/.done/08.ok.
+Spec: docs/specs/08_benchmark_queries.md.
+"""
+
+from __future__ import annotations
+
+import logging
+import sqlite3
+import sys
+from pathlib import Path
+
+import psycopg
+
+from common.pg import get_dsn
+from common.pipeline import (
+	check_dependencies,
+	parse_task_args,
+	process_metrics,
+	remove_stale_marker,
+	write_marker,
+)
+from common.queries import Q1_TRACK_CODE, csv_queries, json_queries, rdf_queries
+from common.results import compare, serialize, sha256_of, sort_rows
+
+logger = logging.getLogger(__name__)
+
+TASK_ID = "08"
+DEPENDS_ON = ["03", "04", "05", "06", "07"]
+
+EXACT_ROW_COUNTS = {1: 1000, 3: 480, 5: 4, 6: 480, 7: 2}  # fixed by corpus volumes
+
+# --- SQL implementations -------------------------------------------------
+# Q2/Q3/Q4/Q7 use the precomputed track_summary (the relational strength);
+# Q5 must scan raw friction_cycles in every format (spec 08).
+
+Q1_SQL = f"""-- Q1: COF vs cycle curve for track {Q1_TRACK_CODE} (point read)
+SELECT fc.cycle, fc.cof
+FROM friction_cycles fc
+JOIN tracks t ON t.track_id = fc.track_id
+WHERE t.track_code = '{Q1_TRACK_CODE}'
+ORDER BY fc.cycle;
+"""
+
+Q2_SQL = """-- Q2: steady-state COF per coupon with mean Au = 10 +/- 0.5 wt%
+SELECT c.coupon_code, t.environment, AVG(s.cof_ss_mean) AS cof_ss
+FROM coupons c
+JOIN tracks t ON t.coupon_id = c.coupon_id
+JOIN track_summary s ON s.track_id = t.track_id
+WHERE c.au_wtpct_mean BETWEEN 9.5 AND 10.5
+GROUP BY c.coupon_code, t.environment
+ORDER BY c.coupon_code;
+"""
+
+Q3_SQL = """-- Q3: hardness vs steady-state COF per friction coupon, across all batches
+SELECT c.coupon_code,
+       b.batch_code,
+       (SELECT AVG(n.hardness_gpa) FROM nanoindentation n WHERE n.coupon_id = c.coupon_id) AS hardness_gpa,
+       AVG(s.cof_ss_mean) AS cof_ss
+FROM coupons c
+JOIN batches b ON b.batch_id = c.batch_id
+JOIN tracks t ON t.coupon_id = c.coupon_id
+JOIN track_summary s ON s.track_id = t.track_id
+GROUP BY c.coupon_id, c.coupon_code, b.batch_code
+ORDER BY c.coupon_code;
+"""
+
+Q4_SQL = """-- Q4: anomaly filter - Dry N2 tracks at 100 mN with cof_ss > 0.20
+SELECT t.track_code, s.cof_ss_mean
+FROM tracks t
+JOIN track_summary s ON s.track_id = t.track_id
+WHERE t.environment = 'dry_n2' AND t.load_mn = 100 AND s.cof_ss_mean > 0.20
+ORDER BY t.track_code;
+"""
+
+# Q5: forced full scan over raw cycles; the run-in algorithm is inlined
+# (docs/rules/db-sql-schema.md section 5). SQLite has no stddev built-in,
+# so the 2-sigma band compares squares: (cof-m)^2 < 4*var  <=>  |cof-m| < 2s.
+Q5_SQLITE = """-- Q5: mean run-in cycles per batch over ALL raw cycle rows (track_summary forbidden)
+WITH tail AS (
+	SELECT track_id, AVG(cof) AS m, AVG(cof * cof) - AVG(cof) * AVG(cof) AS var
+	FROM friction_cycles
+	WHERE cycle >= 501
+	GROUP BY track_id
+),
+run_in AS (
+	SELECT fc.track_id,
+	       COALESCE(MIN(CASE WHEN (fc.cof - tl.m) * (fc.cof - tl.m) < 4 * tl.var THEN fc.cycle END), 500) AS run_in
+	FROM friction_cycles fc
+	JOIN tail tl ON tl.track_id = fc.track_id
+	GROUP BY fc.track_id
+)
+SELECT b.batch_code, AVG(r.run_in) AS avg_run_in_cycles
+FROM run_in r
+JOIN tracks t ON t.track_id = r.track_id
+JOIN coupons c ON c.coupon_id = t.coupon_id
+JOIN batches b ON b.batch_id = c.batch_id
+GROUP BY b.batch_code
+ORDER BY b.batch_code;
+"""
+
+Q5_PG = """-- Q5: mean run-in cycles per batch over ALL raw cycle rows (track_summary forbidden)
+WITH tail AS (
+	SELECT track_id, avg(cof) AS m, stddev_pop(cof) AS s
+	FROM friction_cycles
+	WHERE cycle >= 501
+	GROUP BY track_id
+),
+run_in AS (
+	SELECT fc.track_id,
+	       COALESCE(MIN(fc.cycle) FILTER (WHERE abs(fc.cof - tl.m) < 2 * tl.s), 500) AS run_in
+	FROM friction_cycles fc
+	JOIN tail tl ON tl.track_id = fc.track_id
+	GROUP BY fc.track_id
+)
+SELECT b.batch_code, AVG(r.run_in)::double precision AS avg_run_in_cycles
+FROM run_in r
+JOIN tracks t ON t.track_id = r.track_id
+JOIN coupons c ON c.coupon_id = t.coupon_id
+JOIN batches b ON b.batch_id = c.batch_id
+GROUP BY b.batch_code
+ORDER BY b.batch_code;
+"""
+
+Q6_SQL = """-- Q6: mean wear volume vs load per friction coupon
+SELECT c.coupon_code, t.load_mn, AVG(w.wear_volume_um3) AS wear_volume_um3
+FROM wear w
+JOIN tracks t ON t.track_id = w.track_id
+JOIN coupons c ON c.coupon_id = t.coupon_id
+GROUP BY c.coupon_id, c.coupon_code, t.load_mn
+ORDER BY c.coupon_code;
+"""
+
+Q7_SQL = """-- Q7: Stribeck-style mean COF by (environment, speed*load bucket) past run-in
+SELECT t.environment,
+       CAST(ROUND(t.speed_mm_s * t.load_mn) AS INTEGER) AS speed_load_bucket,
+       AVG(fc.cof) AS mean_cof
+FROM friction_cycles fc
+JOIN tracks t ON t.track_id = fc.track_id
+JOIN track_summary s ON s.track_id = fc.track_id
+WHERE fc.cycle > s.run_in_cycles
+GROUP BY t.environment, speed_load_bucket
+ORDER BY t.environment, speed_load_bucket;
+"""
+
+SQLITE_SQL = {1: Q1_SQL, 2: Q2_SQL, 3: Q3_SQL, 4: Q4_SQL, 5: Q5_SQLITE, 6: Q6_SQL, 7: Q7_SQL}
+PG_SQL = {1: Q1_SQL, 2: Q2_SQL, 3: Q3_SQL, 4: Q4_SQL, 5: Q5_PG, 6: Q6_SQL, 7: Q7_SQL}
+
+RDF_SPARQL_FILES = {
+	1: ("q1",), 2: ("q2",), 3: ("q3_hardness", "q3_cycles"),
+	4: ("q4",), 5: ("q5",), 6: ("q6",), 7: ("q7",),
+}
+
+WRAPPER = '''"""Benchmark query q{n} for the {fmt} format (generated by task 08).
+
+Prints the canonical result rows to stdout at full float precision; the
+benchmark runner (task 09) captures and validates them against
+out/bench/expected/q{n}.rows.csv.
+"""
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[4]
+sys.path.insert(0, str(REPO_ROOT))
+
+from common.queries.{module} import q{n}
+from common.results import print_rows
+
+if __name__ == "__main__":
+	print_rows(q{n}(REPO_ROOT / "out" / {subpath}))
+'''
+
+PY_FORMATS = [
+	("csv", "csv_queries", '"csv"'),
+	("json", "json_queries", '"json"'),
+	("rdf", "rdf_queries", '"rdf" / "oxigraph_store"'),
+]
+
+SQLITE_WRAPPER = '''"""Benchmark query q{n} for the sqlite format (generated by task 08).
+
+Executes the adjacent q{n}.sql against out/sqlite/tribo.db (read-only) and
+prints the canonical result rows to stdout at full float precision; the
+benchmark runner (task 09) captures and validates them against
+out/bench/expected/q{n}.rows.csv.
+"""
+import sqlite3
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[4]
+sys.path.insert(0, str(REPO_ROOT))
+
+from common.results import print_rows
+
+if __name__ == "__main__":
+	sql = Path(__file__).with_suffix(".sql").read_text(encoding="ascii")
+	db = REPO_ROOT / "out" / "sqlite" / "tribo.db"
+	conn = sqlite3.connect(f"file:{{db.as_posix()}}?mode=ro", uri=True)
+	print_rows([tuple(r) for r in conn.execute(sql).fetchall()])
+'''
+
+PG_WRAPPER = '''"""Benchmark query q{n} for the pg format (generated by task 08).
+
+Executes the adjacent q{n}.sql on the PostgreSQL instance addressed by
+TRIBO_PG_DSN and prints the canonical result rows to stdout at full float
+precision; the benchmark runner (task 09) captures and validates them
+against out/bench/expected/q{n}.rows.csv.
+"""
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[4]
+sys.path.insert(0, str(REPO_ROOT))
+
+import psycopg
+
+from common.pg import get_dsn
+from common.results import print_rows
+
+if __name__ == "__main__":
+	sql = Path(__file__).with_suffix(".sql").read_text(encoding="ascii")
+	with psycopg.connect(get_dsn()) as conn:
+		print_rows([tuple(r) for r in conn.execute(sql).fetchall()])
+'''
+
+SQL_WRAPPERS = {"sqlite": SQLITE_WRAPPER, "pg": PG_WRAPPER}
+
+
+def write_query_files(bench_dir: Path) -> int:
+	queries_dir = bench_dir / "queries"
+	written = 0
+	for fmt, module, subpath in PY_FORMATS:
+		fmt_dir = queries_dir / fmt
+		fmt_dir.mkdir(parents=True, exist_ok=True)
+		for n in range(1, 8):
+			text = WRAPPER.format(n=n, fmt=fmt, module=module, subpath=subpath)
+			(fmt_dir / f"q{n}.py").write_text(text, encoding="ascii", newline="\n")
+			written += 1
+	for fmt, sql in (("sqlite", SQLITE_SQL), ("pg", PG_SQL)):
+		fmt_dir = queries_dir / fmt
+		fmt_dir.mkdir(parents=True, exist_ok=True)
+		for n, text in sql.items():
+			(fmt_dir / f"q{n}.sql").write_text(text, encoding="ascii", newline="\n")
+			(fmt_dir / f"q{n}.py").write_text(SQL_WRAPPERS[fmt].format(n=n), encoding="ascii", newline="\n")
+			written += 2
+	rdf_dir = queries_dir / "rdf"
+	for n, keys in RDF_SPARQL_FILES.items():
+		text = "\n\n".join(f"# retrieval query: {key}\n{rdf_queries.SPARQL[key]}" for key in keys)
+		(rdf_dir / f"q{n}.sparql").write_text(text + "\n", encoding="ascii", newline="\n")
+		written += 1
+	logger.info("wrote %d query files under %s", written, queries_dir)
+	return written
+
+
+def run_sqlite(db_path: Path) -> dict[int, list[tuple]]:
+	conn = sqlite3.connect(db_path)
+	try:
+		return {n: [tuple(r) for r in conn.execute(sql).fetchall()] for n, sql in SQLITE_SQL.items()}
+	finally:
+		conn.close()
+
+
+def run_pg(dsn: str) -> dict[int, list[tuple]]:
+	results: dict[int, list[tuple]] = {}
+	with psycopg.connect(dsn) as conn:
+		for n, sql in PG_SQL.items():
+			results[n] = [tuple(r) for r in conn.execute(sql).fetchall()]
+			logger.info("pg q%d: %d rows", n, len(results[n]))
+	return results
+
+
+def main() -> int:
+	args = parse_task_args("Task 08: define benchmark queries and canonical expected results")
+	out_root: Path = args.out_root
+	bench_dir = out_root / "bench"
+	try:
+		check_dependencies(out_root, DEPENDS_ON)
+		remove_stale_marker(out_root, TASK_ID)
+		files_written = write_query_files(bench_dir)
+
+		logger.info("computing canonical results on PostgreSQL")
+		pg_rows = run_pg(get_dsn())
+		logger.info("cross-validating against SQLite")
+		sqlite_rows = run_sqlite(out_root / "sqlite" / "tribo.db")
+		for n in range(1, 8):
+			diff = compare(pg_rows[n], sqlite_rows[n])
+			if diff:
+				raise ValueError(f"q{n}: SQLite disagrees with PostgreSQL: {diff}")
+			exact = EXACT_ROW_COUNTS.get(n)
+			if exact is not None and len(pg_rows[n]) != exact:
+				raise ValueError(f"q{n}: {len(pg_rows[n])} rows, expected {exact}")
+			if not pg_rows[n]:
+				raise ValueError(f"q{n}: empty result set")
+
+		logger.info("spot-checking csv/json/rdf on Q1")
+		for label, rows in (
+			("csv", csv_queries.q1(out_root / "csv")),
+			("json", json_queries.q1(out_root / "json")),
+			("rdf", rdf_queries.q1(out_root / "rdf" / "oxigraph_store")),
+		):
+			diff = compare(pg_rows[1], rows)
+			if diff:
+				raise ValueError(f"q1 via {label} disagrees with canonical: {diff}")
+
+		expected_dir = bench_dir / "expected"
+		expected_dir.mkdir(parents=True, exist_ok=True)
+		checksums = {}
+		for n in range(1, 8):
+			rows = sort_rows(pg_rows[n])
+			(expected_dir / f"q{n}.rows.csv").write_text(serialize(rows), encoding="ascii", newline="\n")
+			checksums[n] = sha256_of(rows)
+			(expected_dir / f"q{n}.sha256").write_text(checksums[n] + "\n", encoding="ascii", newline="\n")
+
+		entries = {"query_files": files_written}
+		for n in range(1, 8):
+			entries[f"q{n}_rows"] = len(pg_rows[n])
+			entries[f"q{n}_sha256"] = checksums[n][:12]
+		entries.update(process_metrics())
+		marker_path = write_marker(out_root, TASK_ID, entries)
+	except Exception:
+		logger.critical("task %s failed", TASK_ID, exc_info=True)
+		return 1
+
+	counts = ", ".join(f"q{n}={len(pg_rows[n])}" for n in range(1, 8))
+	print(f"task 08 ok: {files_written} query files, canonical results validated (pg == sqlite, 1e-9)")
+	print(f"rows: {counts}")
+	print(f"expected: {bench_dir / 'expected'} (q<N>.rows.csv + q<N>.sha256)")
+	print(f"marker: {marker_path}")
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

out/.done/01.ok

@@ -0,0 +1,13 @@
+task: '01'
+status: ok
+config_file: C:/Projects/Public/LabDataStorageEvaluation/out/config/lab_config.yaml
+config_bytes: 5827
+seed: 20260711
+batches: 4
+wafers: 12
+coupons: 588
+friction_coupons: 480
+tracks: 1440
+cycle_rows: 1440000
+loop_points: 2880000
+xrf_points: 235200

out/.done/02.ok

@@ -0,0 +1,6 @@
+task: '02'
+status: ok
+process_flow_file: C:/Projects/Public/LabDataStorageEvaluation/out/report/process_flow.md
+process_flow_bytes: 4205
+diagrams: 6
+diagram_bytes: 2924

out/.done/03.ok

@@ -0,0 +1,11 @@
+task: '03'
+status: ok
+csv_root: C:/Projects/Public/LabDataStorageEvaluation/out/csv
+files: 8718
+total_rows: 4636776
+total_bytes: 148654472
+coupons: 588
+tracks: 1440
+cycle_rows: 1440000
+loop_points: 2880000
+xrf_points: 235200

out/.done/04.ok

@@ -0,0 +1,11 @@
+task: '04'
+status: ok
+full_files: 588
+full_bytes: 311300222
+hybrid_bytes: 1887322
+coupons: 588
+tracks: 1440
+cycle_rows: 1440000
+loop_points: 2880000
+proc_peak_rss_mb: 119.1
+proc_cpu_seconds: 19.1

out/.done/05.ok

@@ -0,0 +1,12 @@
+task: '05'
+status: ok
+db_file: C:/Projects/Public/LabDataStorageEvaluation/out/sqlite/tribo.db
+db_bytes: 123400192
+tables: 15
+total_rows: 4638223
+friction_cycles: 1440000
+friction_loop_points: 2880000
+tracks: 1440
+track_summary: 1440
+proc_peak_rss_mb: 323.0
+proc_cpu_seconds: 19.9

out/.done/06.ok

@@ -0,0 +1,18 @@
+task: '06'
+status: ok
+database: lab_data
+tables: 15
+total_rows: 4638223
+live_bytes: 397467648
+friction_cycles_bytes: 99753984
+friction_loop_points_bytes: 267796480
+copy_seconds: 19.8
+total_seconds: 23.0
+dump: skipped (pg_dump unavailable on this host)
+proc_peak_rss_mb: 58.6
+proc_cpu_seconds: 18.7
+host_cpu_avg_pct: 2.4
+host_cpu_max_pct: 3.7
+host_ram_used_baseline_mb: 617
+host_ram_used_peak_mb: 681
+host_ram_used_delta_mb: 64

out/.done/07.ok

@@ -0,0 +1,13 @@
+task: '07'
+status: ok
+triples: 17419713
+nt_gz_bytes: 126487650
+ttl_bytes: 596401229
+store_bytes: 1739414085
+store_load_seconds: 33.6
+coupons: 588
+tracks: 1440
+cycle_nodes: 1440000
+loop_nodes: 2880000
+proc_peak_rss_mb: 2185.2
+proc_cpu_seconds: 151.8

out/.done/08.ok

@@ -0,0 +1,19 @@
+task: '08'
+status: ok
+query_files: 56
+q1_rows: 1000
+q1_sha256: bb6d871275d0
+q2_rows: 33
+q2_sha256: 62ca30e729a8
+q3_rows: 480
+q3_sha256: a3a73414201f
+q4_rows: 469
+q4_sha256: 38281dcf7896
+q5_rows: 4
+q5_sha256: f3ee32dbc981
+q6_rows: 480
+q6_sha256: 9cc28d51ee35
+q7_rows: 2
+q7_sha256: 1991f8bf2217
+proc_peak_rss_mb: 61.1
+proc_cpu_seconds: 1.7

out/.done/09.ok

@@ -0,0 +1,14 @@
+task: '09'
+status: ok
+cells_total: 35
+cells_ok: 35
+cells_invalid: 0
+cells_timeout: 0
+cells_error: 0
+session_wall_s: 915.5
+cold_cache_pass: skipped (Windows host)
+session_host_cpu_avg_pct: 0.4
+session_host_cpu_max_pct: 9.2
+session_host_ram_used_baseline_mb: 909
+session_host_ram_used_peak_mb: 914
+session_host_ram_used_delta_mb: 5

out/.done/10.ok

@@ -0,0 +1,16 @@
+task: '10'
+status: ok
+extrapolation_rows: 105
+sizing_rows: 15
+flags_ok: 53
+flags_impractical: 35
+flags_fail: 17
+pg_index_share: 0.341
+pg_index_mb: 128.5
+prices_as_of: 2026-07-11
+note_pandas: pandas CSV variant not measured; its RAM is O(n) - infeasible above ~10 GB (spec 10)
+coeff_csv: 1.0
+coeff_json: 2.094
+coeff_sqlite: 0.83
+coeff_pg: 2.674
+coeff_rdf: 11.701

out/.done/11.ok

@@ -0,0 +1,12 @@
+task: '11'
+status: ok
+charts: 14
+tables: 7
+report_bytes: 16091
+winner_600gb: pg
+prices_as_of: 2026-07-11
+score600_csv: 28.9
+score600_json: 24.0
+score600_sqlite: 47.0
+score600_pg: 58.6
+score600_rdf: 0.9

out/config/hw_prices.yaml

@@ -0,0 +1,18 @@
+# Hardware cost parameters for extrapolation (task 10).
+# Street prices retrieved 2026-07-11 from the sources below (server DRAM is
+# in a documented 2025-2026 price surge - see market_context). The cost
+# model buys whole components: 64 GB RDIMM modules and 7.68 TB NVMe drives.
+# Operator-editable; re-run extrapolate.py after changes (code-config-yaml).
+as_of: "2026-07-11"
+ram_module_gb: 64
+ram_module_usd: 1887.0       # A-Tech 64GB (2x32GB) DDR5-5600 ECC RDIMM
+nvme_drive_tb: 7.68
+nvme_drive_usd: 3995.0       # Cloud Ninjas NEW 7.68TB NVMe U.2 1DWPD (Dell 14-16G)
+server_16_core_usd: 4618.0   # Supermicro AS-1015CS-TNR 1U (EPYC 9004/9005), Broadberry starting config
+server_32_core_usd: 6272.0   # Supermicro AS-1115CS-TNR 1U, Broadberry starting config
+extra_node_usd: 4618.0       # one additional entry 1U node (AS-1015CS-TNR chassis)
+sources:
+  ram: https://atechmemory.com/collections/ddr5-memory-ram
+  nvme: https://cloudninjas.com/products/new-7-68tb-nvme-u-2-1dwpd-sie-2-5-enterprise-solid-state-drive-for-14th-15th-16th-gen-dell
+  servers: https://www.broadberry.com/amd-epyc-9004-supermicro-servers
+  market_context: https://www.techpowerup.com/342331/server-dram-pricing-jumps-50-only-70-of-orders-getting-filled

out/config/lab_config.yaml

@@ -0,0 +1,246 @@
+# Generated by make_lab_config.py (task 01) from docs/specs/01_lab_configuration.md.
+# Do not hand-edit: regenerate via `python make_lab_config.py`.
+project: LabDataStorageEvaluation
+description: Simulated tribology laboratory (Sandia Pt-Au LDRD context)
+spec: docs/specs/01_lab_configuration.md
+seed: 20260711
+material_system:
+  substrate:
+    material: Ti-6Al-4V
+    dimensions_mm:
+    - 10
+    - 10
+    - 3
+  adhesion_layer:
+    material: Cr
+    process: sputtered
+  coating:
+    material: Pt-Au
+    deposition: composition gradient across each wafer
+    thickness_um_range:
+    - 0.3
+    - 1.1
+instruments:
+- instrument_id: rapid
+  role: friction
+  name: RAPID custom high-throughput parallelized 6-probe tribometer
+  data_produced: COF vs cycle per track; test conditions
+- instrument_id: ti980
+  role: nanoindentation
+  name: Bruker TI980 TriboIndenter
+  data_produced: hardness, reduced modulus (25 indents per coupon)
+- instrument_id: m4_tornado
+  role: composition
+  name: Bruker M4 Tornado micro-XRF
+  data_produced: Pt/Au wt% map per coupon (20x20 grid)
+- instrument_id: pvd200
+  role: deposition
+  name: Kurt J. Lesker PVD 200 sputter-down
+  data_produced: batch deposition parameters
+- instrument_id: afm
+  role: surface_roughness
+  name: AFM
+  data_produced: topography image metadata + Ra/Rq per coupon
+- instrument_id: profilometer
+  role: film_thickness
+  name: Optical profilometry
+  data_produced: thickness map per coupon (10x10 grid)
+- instrument_id: simtra
+  role: simulation
+  name: SIMTRA sputter transport Monte-Carlo
+  data_produced: deposition atom-energy / composition profiles per deposition run
+deposition_matrix:
+- batch_code: B721
+  pt_gun_tilt_deg: 20
+  au_gun_tilt_deg: 0
+  pt_power_W: 150
+  au_power_W: 50
+  pt_discharge_V: 432
+  au_discharge_V: 311
+- batch_code: B722
+  pt_gun_tilt_deg: 20
+  au_gun_tilt_deg: 20
+  pt_power_W: 100
+  au_power_W: 100
+  pt_discharge_V: 399
+  au_discharge_V: 352
+- batch_code: B723
+  pt_gun_tilt_deg: 20
+  au_gun_tilt_deg: 20
+  pt_power_W: 150
+  au_power_W: 50
+  pt_discharge_V: 430
+  au_discharge_V: 311
+- batch_code: B724
+  pt_gun_tilt_deg: 0
+  au_gun_tilt_deg: 20
+  pt_power_W: 50
+  au_power_W: 150
+  pt_discharge_V: 376
+  au_discharge_V: 340
+hierarchy:
+  batches: 4
+  wafers_per_batch: 3
+  coupon_grid_per_wafer:
+  - 7
+  - 7
+  coupons_per_wafer: 49
+friction_assignment:
+  friction_coupons_total: 480
+  reserve_coupons_total: 108
+  runs: 4
+  coupons_per_run: 120
+  plates_per_run: 5
+  probes_per_plate: 6
+  coupons_per_probe_square: 4
+  tracks_per_friction_coupon: 3
+  counterfaces_per_holder: 3
+  fresh_counterface_per_track: true
+  reserve_grid_positions:
+  - 1
+  - 4
+  - 7
+  - 22
+  - 25
+  - 28
+  - 43
+  - 46
+  - 49
+volumes:
+  coupons_total: 588
+  tracks_total: 1440
+  cycles_per_track: 1000
+  cycle_rows_total: 1440000
+  loop_every_n_cycles: 100
+  loops_per_track: 10
+  loop_points_per_loop: 200
+  loop_points_total: 2880000
+  xrf_grid:
+  - 20
+  - 20
+  xrf_points_per_coupon: 400
+  xrf_points_total: 235200
+  profilometry_grid:
+  - 10
+  - 10
+  profilometry_points_per_coupon: 100
+  nanoindentation_indents_per_coupon: 25
+  simtra_rows_per_batch: 1000
+runs:
+- run_code: R1
+  batch_code: B721
+  environment: lab_air
+  date: '2026-04-06'
+  operator: A. Reyes
+- run_code: R2
+  batch_code: B722
+  environment: lab_air
+  date: '2026-04-13'
+  operator: K. Patel
+- run_code: R3
+  batch_code: B723
+  environment: dry_n2
+  date: '2026-04-20'
+  operator: A. Reyes
+- run_code: R4
+  batch_code: B724
+  environment: dry_n2
+  date: '2026-04-27'
+  operator: M. Novak
+schedule:
+  deposition_start_date: '2026-03-02'
+  batch_interval_days: 7
+  characterization_lag_days: 3
+  run_start_time: 09:00:00
+  track_interval_s: 2100
+test_conditions:
+  normal_load_mN: 100
+  stroke_mm: 1.0
+  speed_mm_s: 1.0
+  counterface:
+    material: ruby (Al2O3)
+    diameter_mm: 3.175
+  rh_pct:
+    lab_air: 45
+    dry_n2: 2
+  temperature_C: 23
+  temperature_tolerance_C: 1
+physical_models:
+  au_gradient:
+    description: linear Au wt% gradient along wafer x, batch-dependent center
+    batch_center_wtpct:
+      B721: 8
+      B722: 25
+      B723: 10
+      B724: 60
+    wafer_span_wtpct: 5
+    xrf_noise_sd_wtpct: 0.3
+  cof_run_in:
+    formula: cof(c) = cof_ss + (cof_0 - cof_ss) * exp(-c / tau) + N(0, sigma)
+    cof_0_range:
+    - 0.35
+    - 0.5
+    tau_cycles_range:
+    - 100
+    - 400
+    noise_sd: 0.01
+  cof_steady_state:
+    formula: cof_ss = intercept + slope_per_au_wtpct * au_wtpct_mean, clamped to >= floor
+    lab_air:
+      intercept: 0.32
+      slope_per_au_wtpct: -0.0015
+    dry_n2:
+      intercept: 0.24
+      slope_per_au_wtpct: -0.0012
+    floor: 0.12
+  hardness_GPa:
+    formula: H = intercept + slope_per_au_wtpct * au_wtpct_mean + N(0, noise_sd)
+    intercept: 8.5
+    slope_per_au_wtpct: -0.05
+    noise_sd: 0.25
+  reduced_modulus_GPa:
+    formula: Er = intercept + slope_per_au_wtpct * au_wtpct_mean + N(0, noise_sd)
+    intercept: 190
+    slope_per_au_wtpct: -0.6
+    noise_sd: 4
+  roughness_ra_nm:
+    distribution: lognormal
+    median_nm: 5
+    sigma_log: 0.3
+    rq_over_ra: 1.25
+  thickness_um:
+    profile: radial parabolic across wafer plus noise
+    center_um: 1.05
+    edge_um: 0.35
+    noise_sd_um: 0.02
+    range_um:
+    - 0.3
+    - 1.1
+  wear_archard:
+    formula: V = k * F * s; k = k0 * (1 + slope_per_au_wtpct * au_wtpct_mean), lognormal noise, clamped
+      to >= k_floor
+    k0_mm3_per_N_m:
+      lab_air: 1.5e-07
+      dry_n2: 6.0e-08
+    slope_per_au_wtpct:
+      lab_air: -0.008
+      dry_n2: -0.01
+    k_floor_mm3_per_N_m: 1.0e-09
+    noise_sigma_log: 0.15
+  friction_loop:
+    force_noise_sd_mN: 0.3
+  nanoindentation:
+    max_load_mN: 10
+    indent_grid:
+    - 5
+    - 5
+    grid_pitch_um: 20
+  simtra:
+    angle_deg_range:
+    - 0
+    - 90
+    energy_eV_max: 50
+    surface_binding_energy_eV:
+      pt: 5.84
+      au: 3.81
+    flux_per_W: 0.02

out/report/REPORT.md

@@ -0,0 +1,318 @@
+# Laboratory Data Storage Format Evaluation - Final Report
+
+Generated by task 11 (report.py) from the measured benchmark matrix (task 09) and the extrapolation model (task 10). Every number at scales beyond the 150 MB corpus is a PROJECTION and is labeled as such; scaling assumptions are listed in section 6.
+
+## 1. Executive summary
+
+**Recommendation: PostgreSQL as the system of record.** The measured matrix and the projections confirm the expected outcome:
+
+- **PostgreSQL** is the only format whose seven benchmark queries stay below the 1-hour practicality bound at every projected scale (600 GB, 1.2 TB, 6 TB); at 6 TB it needs one 16-core node with 551 GB RAM at an estimated $33,586 (bill of materials, prices as of 2026-07-11). Partitioned parallel scans and the track_summary materialized view are the scaling levers.
+- **CSV** remains the canonical raw archive: byte-reproducible, instrument-native, and the best compressor (tar.gz ~2.8x); its full scans exceed 1 h beyond ~600 GB, so it is an archive, not a query layer.
+- **SQLite** is an excellent single-user store up to ~600 GB (all queries OK), then its single-threaded scans degrade (Q5/Q7 IMPRACTICAL at 1.2 TB).
+- **Hybrid JSON-LD** (metadata + sourceFile links, 1.8 MiB at 150 MB corpus) is the exchange format; the FULL variant re-reads the whole corpus per scan and becomes impractical past 600 GB.
+- **RDF (materialized triplestore)** fails at scale: 11.7x storage blow-up, 5 of 7 queries FAIL at 6 TB, and the hot set needs 13.7 TB RAM across 14 nodes (~$527,732 projected). Semantic access should be a virtual layer (e.g. Ontop OBDA) over PostgreSQL instead.
+
+## 2. Methodology snapshot
+
+35 cells (Q1-Q7 x 5 formats); per cell 1 warm-up + 3 measured runs in isolated subprocesses sampled with psutil at 50 ms; randomized cell order; 30-min timeout; every run validated against canonical results (PostgreSQL cross-validated with SQLite, tolerance 1e-9). Medians are reported; raw runs and min/max live in out/bench/results_raw.csv. The cold-cache pass is skipped on this Windows host and marked as such. PostgreSQL runs on a remote host; its wall times include the LAN round-trip, and server-side metrics come from pg_stat_statements and Prometheus (docs/rules/bench-methodology.md section 6).
+
+## 3. Environment
+
+```
+benchmark_host_os: Windows-2025Server-10.0.26100-SP0
+benchmark_host_cpu: Intel64 Family 6 Model 141 Stepping 1, GenuineIntel
+benchmark_host_cores_logical: 4
+benchmark_host_cores_physical: 4
+benchmark_host_ram_gib: 8.0
+python: 3.12.10
+benchmark_host_disks: VMware Virtual NVMe Disk / SSD / 100 GB
+sqlite: 3.49.1
+psycopg: 3.3.4
+pyoxigraph: 0.5.9
+ijson: 3.5.1
+psutil: 7.2.2
+postgresql_server: PostgreSQL 16.14 (Ubuntu 16.14-0ubuntu0.24.04.1) on x86_64-pc-linux-gnu, compiled by gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0, 64-bit
+postgresql_host: 192.168.10.73 (remote; 4 cores / 3.8 GiB, see .done/06.ok)
+cold_cache_pass: skipped (Windows host, no page-cache drop - bench-methodology section 6)
+pg_backend_sampling: not possible for a remote server; Prometheus windows + pg_stat_statements deltas instead
+```
+
+## 4. Storage footprint
+
+| format | measured | vs CSV | archival (compressed) | projected @600 GB | projected @1.2 TB | projected @6 TB |
+|---|---|---|---|---|---|---|
+| CSV | 149 MB | 1.00x | 53 MB | 600 GB | 1.20 TB | 6.00 TB |
+| JSON-LD | 311 MB | 2.09x | 57 MB | 1.26 TB | 2.51 TB | 13 TB |
+| SQLite | 123 MB | 0.83x | 55 MB | 498 GB | 996 GB | 4.98 TB |
+| PostgreSQL | 397 MB | 2.67x | 53 MB | 1.60 TB | 3.21 TB | 16 TB |
+| RDF triplestore | 1.74 GB | 11.70x | 126 MB | 7.02 TB | 14 TB | 70 TB |
+
+![C1 - disk footprint](charts/c1_disk_footprint.png)
+
+## 5. Measured performance (150 MB corpus)
+
+| format | query | wall time | peak RSS | CPU util | read |
+|---|---|---|---|---|---|
+| CSV | q1 | 0.06 s | 4 MB | 0% | 0.0 MB |
+| CSV | q2 | 0.17 s | 20 MB | 16% | 2.5 MB |
+| CSV | q3 | 1.15 s | 21 MB | 24% | 31.8 MB |
+| CSV | q4 | 0.71 s | 21 MB | 22% | 16.8 MB |
+| CSV | q5 | 1.09 s | 21 MB | 23% | 31.8 MB |
+| CSV | q6 | 0.28 s | 20 MB | 17% | 1.2 MB |
+| CSV | q7 | 1.27 s | 21 MB | 23% | 32.5 MB |
+| JSON-LD | q1 | 0.11 s | 22 MB | 10% | 1.4 MB |
+| JSON-LD | q2 | 1.48 s | 24 MB | 24% | 53.1 MB |
+| JSON-LD | q3 | 9.94 s | 24 MB | 25% | 297.6 MB |
+| JSON-LD | q4 | 9.92 s | 24 MB | 25% | 297.8 MB |
+| JSON-LD | q5 | 9.96 s | 24 MB | 25% | 297.6 MB |
+| JSON-LD | q6 | 9.79 s | 24 MB | 24% | 297.8 MB |
+| JSON-LD | q7 | 10 s | 24 MB | 24% | 297.9 MB |
+| SQLite | q1 | 0.06 s | 4 MB | 0% | 0.0 MB |
+| SQLite | q2 | 0.06 s | 4 MB | 0% | 0.0 MB |
+| SQLite | q3 | 0.11 s | 4 MB | 0% | 0.0 MB |
+| SQLite | q4 | 0.06 s | 4 MB | 0% | 0.0 MB |
+| SQLite | q5 | 0.77 s | 25 MB | 23% | 117.9 MB |
+| SQLite | q6 | 0.06 s | 4 MB | 0% | 0.0 MB |
+| SQLite | q7 | 0.71 s | 25 MB | 23% | 93.0 MB |
+| PostgreSQL | q1 | 0.22 s | 42 MB | 18% | 4.1 MB |
+| PostgreSQL | q2 | 0.22 s | 42 MB | 20% | 4.1 MB |
+| PostgreSQL | q3 | 0.22 s | 42 MB | 18% | 4.1 MB |
+| PostgreSQL | q4 | 0.22 s | 42 MB | 18% | 4.1 MB |
+| PostgreSQL | q5 | 0.44 s | 42 MB | 10% | 4.1 MB |
+| PostgreSQL | q6 | 0.22 s | 42 MB | 19% | 4.1 MB |
+| PostgreSQL | q7 | 0.33 s | 42 MB | 12% | 4.1 MB |
+| RDF triplestore | q1 | 0.11 s | 32 MB | 17% | 3.6 MB |
+| RDF triplestore | q2 | 2.81 s | 85 MB | 24% | 375.6 MB |
+| RDF triplestore | q3 | 46 s | 264 MB | 25% | 5445.1 MB |
+| RDF triplestore | q4 | 20 s | 168 MB | 25% | 2694.6 MB |
+| RDF triplestore | q5 | 60 s | 264 MB | 25% | 5390.5 MB |
+| RDF triplestore | q6 | 0.17 s | 46 MB | 16% | 15.9 MB |
+| RDF triplestore | q7 | 41 s | 264 MB | 25% | 5380.5 MB |
+
+Sub-interval note: cells faster than the 50 ms sampling cadence (fast SQLite/CSV point reads) under-report client RSS/CPU; wall times are exact.
+
+![C2 - RAM requirement](charts/c2_ram_requirement.png)
+
+Per-query charts: [Q1](charts/c3_q1_wall_time.png), [Q2](charts/c3_q2_wall_time.png), [Q3](charts/c3_q3_wall_time.png), [Q4](charts/c3_q4_wall_time.png), [Q5](charts/c3_q5_wall_time.png), [Q6](charts/c3_q6_wall_time.png), [Q7](charts/c3_q7_wall_time.png)
+
+## 6. Projections (600 GB / 1.2 TB / 6 TB)
+
+All values in this section are projected, never measured. Laws per access pattern (spec 10): flat / index-depth log / linear result set / linear x depth / full scan / parallel partitioned scan; constants calibrated on the measured point after subtracting the per-format harness floor. Flags: IMPRACTICAL > 1 h, FAIL > 24 h.
+
+Projected wall times at 600 GB:
+
+| format | query | projected wall time | flag |
+|---|---|---|---|
+| CSV | q1 | 0.06 s | OK |
+| CSV | q2 | 7.1 min | OK |
+| CSV | q3 | 73.5 min | IMPRACTICAL |
+| CSV | q4 | 44.1 min | OK |
+| CSV | q5 | 69.6 min | IMPRACTICAL |
+| CSV | q6 | 14.5 min | OK |
+| CSV | q7 | 81.6 min | IMPRACTICAL |
+| JSON-LD | q1 | 0.11 s | OK |
+| JSON-LD | q2 | 92.0 min | IMPRACTICAL |
+| JSON-LD | q3 | 11.0 h | IMPRACTICAL |
+| JSON-LD | q4 | 11.0 h | IMPRACTICAL |
+| JSON-LD | q5 | 11.0 h | IMPRACTICAL |
+| JSON-LD | q6 | 10.9 h | IMPRACTICAL |
+| JSON-LD | q7 | 11.2 h | IMPRACTICAL |
+| SQLite | q1 | 0.06 s | OK |
+| SQLite | q2 | 1.27 s | OK |
+| SQLite | q3 | 5.7 min | OK |
+| SQLite | q4 | 0.06 s | OK |
+| SQLite | q5 | 47.5 min | OK |
+| SQLite | q6 | 5.18 s | OK |
+| SQLite | q7 | 43.7 min | OK |
+| PostgreSQL | q1 | 0.22 s | OK |
+| PostgreSQL | q2 | 0.22 s | OK |
+| PostgreSQL | q3 | 3.42 s | OK |
+| PostgreSQL | q4 | 11 s | OK |
+| PostgreSQL | q5 | 3.7 min | OK |
+| PostgreSQL | q6 | 4.06 s | OK |
+| PostgreSQL | q7 | 115 s | OK |
+| RDF triplestore | q1 | 0.11 s | OK |
+| RDF triplestore | q2 | 3.0 h | IMPRACTICAL |
+| RDF triplestore | q3 | 2.1 d | FAIL |
+| RDF triplestore | q4 | 22.1 h | IMPRACTICAL |
+| RDF triplestore | q5 | 2.8 d | FAIL |
+| RDF triplestore | q6 | 5.9 min | OK |
+| RDF triplestore | q7 | 45.3 h | FAIL |
+
+Degradation by data size: [point read](charts/c4_degradation_point_read.png), [indexed / join](charts/c4_degradation_indexed.png), [full scan](charts/c4_degradation_full_scan.png)
+
+## 7. Hardware sizing and cost at 600 GB (projected)
+
+Bill-of-materials costing with street prices as of **2026-07-11** (sources below). Every configuration buys whole components: 64 GB DDR5 ECC RDIMM modules, 7.68 TB enterprise NVMe U.2 drives, and a priced 1U chassis; disk capacity includes 30% free-space headroom.
+
+| format | configuration | chassis | RAM | disk | extra nodes | total |
+|---|---|---|---|---|---|---|
+| CSV | 1 x 1U Supermicro AS-1015CS-TNR (16 cores); 1 x 64 GB DDR5 ECC RDIMM; 1 x 7.68 TB NVMe U.2 | $4,618 | $1,887 | $3,995 | $0 | $10,500 |
+| JSON-LD | 1 x 1U Supermicro AS-1015CS-TNR (16 cores); 1 x 64 GB DDR5 ECC RDIMM; 1 x 7.68 TB NVMe U.2 | $4,618 | $1,887 | $3,995 | $0 | $10,500 |
+| SQLite | 1 x 1U Supermicro AS-1015CS-TNR (16 cores); 1 x 64 GB DDR5 ECC RDIMM; 1 x 7.68 TB NVMe U.2 | $4,618 | $1,887 | $3,995 | $0 | $10,500 |
+| PostgreSQL | 1 x 1U Supermicro AS-1015CS-TNR (16 cores); 1 x 64 GB DDR5 ECC RDIMM; 1 x 7.68 TB NVMe U.2 | $4,618 | $1,887 | $3,995 | $0 | $10,500 |
+| RDF triplestore | 2 x 1U Supermicro AS-1015CS-TNR (16 cores); 22 x 64 GB DDR5 ECC RDIMM; 2 x 7.68 TB NVMe U.2 | $4,618 | $41,514 | $7,990 | $4,618 | $58,740 |
+
+Component prices (retrieved 2026-07-11):
+
+- 64 GB DDR5-5600 ECC RDIMM (A-Tech, 2x32GB kit): $1,887 - <https://atechmemory.com/collections/ddr5-memory-ram>
+- 7.68 TB NVMe U.2 1DWPD enterprise SSD (Dell 14-16G compatible, Cloud Ninjas): $3,995 - <https://cloudninjas.com/products/new-7-68tb-nvme-u-2-1dwpd-sie-2-5-enterprise-solid-state-drive-for-14th-15th-16th-gen-dell>
+- 1U AMD EPYC 9004/9005 chassis (Broadberry starting configurations): Supermicro AS-1015CS-TNR $4,618, Supermicro AS-1115CS-TNR $6,272 - <https://www.broadberry.com/amd-epyc-9004-supermicro-servers>
+- Market context - server DRAM is in a documented 2025-2026 price surge: <https://www.techpowerup.com/342331/server-dram-pricing-jumps-50-only-70-of-orders-getting-filled>
+
+Full sizing at 1.2 TB and 6 TB: out/bench/hardware_sizing.csv. RDF exceeds the 1 TB single-node RAM ceiling already at 600 GB and reaches 14 nodes at 6 TB.
+
+![C6 - cost vs performance](charts/c6_cost_vs_performance.png)
+
+## 8. Weighted scoring
+
+Subscores are 10 x best/value per metric; weights: search x5, RAM economy x2, disk economy x1; max 80. Search uses the inverse geometric mean of Q1-Q7 (FAIL at 600 GB zeroes the search subscore). Left: measured scale; right: projected 600 GB.
+
+Measured scale (150 MB):
+
+| format | search (x5) | RAM (x2) | disk (x1) | total (max 80) |
+|---|---|---|---|---|
+| SQLite | 50.0 | 16.9 | 10.0 | 76.9 |
+| CSV | 15.5 | 20.0 | 8.3 | 43.8 |
+| PostgreSQL | 26.2 | 10.0 | 3.1 | 39.3 |
+| JSON-LD | 1.7 | 17.2 | 4.0 | 22.9 |
+| RDF triplestore | 1.3 | 1.6 | 0.7 | 3.6 |
+
+Projected at 600 GB:
+
+| format | search (x5) | RAM (x2) | disk (x1) | total (max 80) |
+|---|---|---|---|---|
+| PostgreSQL | 50.0 | 5.5 | 3.1 | 58.6 |
+| SQLite | 21.8 | 15.2 | 10.0 | 47.0 |
+| CSV | 0.6 | 20.0 | 8.3 | 28.9 |
+| JSON-LD | 0.1 | 20.0 | 4.0 | 24.0 |
+| RDF triplestore | 0.0 | 0.2 | 0.7 | 0.9 |
+
+![C5 - weighted score](charts/c5_weighted_score.png)
+
+## 9. Use-case mapping
+
+| Use case | Recommended format(s) |
+|---|---|
+| Interactive analysis (joins, aggregations) | PostgreSQL; SQLite acceptable single-user up to ~600 GB |
+| Report generation (repeated summaries) | PostgreSQL (track_summary materialized view) |
+| Search / filtering | PostgreSQL or SQLite (indexed); flat formats need full scans |
+| Archiving | CSV tree + tar.gz (canonical raw) plus pg_dump of the system of record |
+| Inter-lab exchange | Hybrid JSON-LD (metadata + sourceFile links to CSV) |
+| Semantic / ontology queries | Virtual RDF layer over PostgreSQL (e.g. Ontop OBDA), not a materialized triplestore |
+
+## 10. Limitations
+
+- Single-node measurements on one Windows host + one remote PostgreSQL host; no cluster or cloud variance.
+- The corpus is simulated (physically plausible models, fixed seed 20260711); real instrument data may have different value distributions, though volumes and shapes match the lab.
+- Extrapolation uses analytic laws calibrated on a single 150 MB point; no intermediate-scale validation runs were performed.
+- Client wall times include a per-format harness floor (interpreter start, imports, connection); the floor is assumed constant across scales. The cheapest cell of each format therefore projects flat.
+- The cold-cache matrix pass is skipped (no page-cache drop on Windows); all measured runs are warm-cache.
+- The pandas CSV variant was not measured; its RAM is O(n) and it is infeasible above roughly 10 GB (spec 10).
+- RDF numbers reflect the compact modeling and a client that derives run-in / steady-state from raw cycles; SPARQL-side aggregation engines could shift (not remove) the scan penalty.
+- Hardware prices are spot street prices retrieved 2026-07-11 during a documented server-DRAM price surge; re-cost by editing out/config/hw_prices.yaml and re-running tasks 10-11.
+
+## Appendix A - process flow diagrams (task 02)
+
+### D1 - Coupon assembly
+
+```mermaid
+flowchart LR
+    A([Ti-6Al-4V Base 10x10x3 mm]) --> B[Cleaning]
+    B --> B1([Rinsed in deionized water])
+    B1 --> B2([Sonicated in cleaning solution])
+    B --> C[Smearing Adhesive]
+    C --> C1([Cr adhesive coating])
+    C --> D[Vapor Deposition System<br/>Kurt J. Lesker PVD 200]
+    D --> D1([PtAu sputter coating, gradient])
+    D --> E([Test Coupon])
+```
+
+### D2 - Characterization and batch assembly
+
+```mermaid
+flowchart LR
+    TC([Test Coupon]) --> S[SIMTRA simulation<br/>composition / atom energies]
+    S --> OP[Optical profilometry<br/>film thickness 0.3-1.1 um]
+    OP --> TW[Assemble Test Wafer<br/>x49 coupons]
+    TW --> TB[Assemble Test Batch<br/>x3 wafers]
+    TB --> B721[B721<br/>Pt 150 W / Au 50 W]
+    TB --> B722[B722<br/>Pt 100 W / Au 100 W]
+    TB --> B723[B723<br/>Pt 150 W / Au 50 W]
+    TB --> B724[B724<br/>Pt 50 W / Au 150 W]
+```
+
+### D3 - Testing tree
+
+```mermaid
+flowchart TD
+    TC([Test Coupon]) --> F[Friction<br/>RAPID 6-probe tribometer]
+    F --> FA([Lab Air -> COF dataset mu_normal])
+    F --> FN([Dry N2 -> COF dataset mu_dry_nit])
+    TC --> NI[Nanoindentation<br/>Bruker TI980]
+    NI --> NIR([Hardness, reduced modulus<br/>25 indents per coupon])
+    TC --> AF[AFM]
+    AF --> AFR([Topography -> Ra / Rq])
+    TC --> XR[micro-XRF<br/>Bruker M4 Tornado]
+    XR --> XRR([Pt/Au wt% map, 20x20 grid])
+```
+
+### D4 - Tribometer session sequence
+
+```mermaid
+sequenceDiagram
+    actor Op as Operator
+    participant Prep as Sample Prep
+    participant Plan as Excel Test Plan
+    participant SW as Control Software
+    participant TR as RAPID Tribometer
+
+    Op->>Prep: Request test
+    Op->>Prep: Get samples
+    Op->>Prep: Load samples onto plates
+    Op->>Prep: Load ball holders (3 counterfaces per holder)
+    Op->>Plan: Create test plan<br/>(Sample Plate, Plate Location, Sample ID, Save Location,<br/>Folder Name, X/Y/Z Offset, Load, Iterations)
+    Op->>TR: Transfer plates to tribometer
+    Op->>TR: Set up platter
+    Op->>SW: Activate software
+    Op->>SW: Load plan
+    SW->>TR: Activate tribometer
+    TR-->>SW: Run reciprocating tests (execution loop, see D5)
+    Op->>SW: Software stop
+    SW-->>Op: Save avg files
+    Op->>TR: Remove plates and platters
+    Op->>SW: Close software
+    Op->>TR: Pull equipment out
+```
+
+### D5 - Execution loop
+
+```mermaid
+flowchart TD
+    L[Load all counterfaces] --> P{5 plates}
+    P -->|for each plate| PR{6 probes in parallel}
+    PR -->|for each probe| C{4 coupons on coupon square}
+    C -->|for each coupon| T[Draw track]
+    T --> RC[Rotate counterface - fresh ball per track]
+    RC -->|3 tracks per coupon| T
+    T --> DONE([120 coupons x 3 tracks = 360 tracks per run])
+```
+
+### D6 - Data hierarchy
+
+```mermaid
+erDiagram
+    DEPOSITION_RUN ||--|| SIMTRA_PROFILE : produces
+    DEPOSITION_RUN ||--|| BATCH : creates
+    BATCH ||--|{ WAFER : contains
+    WAFER ||--|{ COUPON : contains
+    COUPON ||--|| XRF_MAP : has
+    COUPON ||--|| NANOINDENTATION : has
+    COUPON ||--|| AFM : has
+    COUPON ||--|| PROFILOMETRY : has
+    COUPON ||--o{ TRACK : "friction coupons only"
+    TRACK ||--|{ CYCLE : has
+    TRACK ||--|{ LOOP_POINT : has
+    TRACK ||--|| WEAR : has
+```
+

out/report/diagrams/D1.mermaid

@@ -0,0 +1,9 @@
+flowchart LR
+    A([Ti-6Al-4V Base 10x10x3 mm]) --> B[Cleaning]
+    B --> B1([Rinsed in deionized water])
+    B1 --> B2([Sonicated in cleaning solution])
+    B --> C[Smearing Adhesive]
+    C --> C1([Cr adhesive coating])
+    C --> D[Vapor Deposition System<br/>Kurt J. Lesker PVD 200]
+    D --> D1([PtAu sputter coating, gradient])
+    D --> E([Test Coupon])

out/report/diagrams/D2.mermaid

@@ -0,0 +1,9 @@
+flowchart LR
+    TC([Test Coupon]) --> S[SIMTRA simulation<br/>composition / atom energies]
+    S --> OP[Optical profilometry<br/>film thickness 0.3-1.1 um]
+    OP --> TW[Assemble Test Wafer<br/>x49 coupons]
+    TW --> TB[Assemble Test Batch<br/>x3 wafers]
+    TB --> B721[B721<br/>Pt 150 W / Au 50 W]
+    TB --> B722[B722<br/>Pt 100 W / Au 100 W]
+    TB --> B723[B723<br/>Pt 150 W / Au 50 W]
+    TB --> B724[B724<br/>Pt 50 W / Au 150 W]

out/report/diagrams/D3.mermaid

@@ -0,0 +1,10 @@
+flowchart TD
+    TC([Test Coupon]) --> F[Friction<br/>RAPID 6-probe tribometer]
+    F --> FA([Lab Air -> COF dataset mu_normal])
+    F --> FN([Dry N2 -> COF dataset mu_dry_nit])
+    TC --> NI[Nanoindentation<br/>Bruker TI980]
+    NI --> NIR([Hardness, reduced modulus<br/>25 indents per coupon])
+    TC --> AF[AFM]
+    AF --> AFR([Topography -> Ra / Rq])
+    TC --> XR[micro-XRF<br/>Bruker M4 Tornado]
+    XR --> XRR([Pt/Au wt% map, 20x20 grid])

out/report/diagrams/D4.mermaid

@@ -0,0 +1,23 @@
+sequenceDiagram
+    actor Op as Operator
+    participant Prep as Sample Prep
+    participant Plan as Excel Test Plan
+    participant SW as Control Software
+    participant TR as RAPID Tribometer
+
+    Op->>Prep: Request test
+    Op->>Prep: Get samples
+    Op->>Prep: Load samples onto plates
+    Op->>Prep: Load ball holders (3 counterfaces per holder)
+    Op->>Plan: Create test plan<br/>(Sample Plate, Plate Location, Sample ID, Save Location,<br/>Folder Name, X/Y/Z Offset, Load, Iterations)
+    Op->>TR: Transfer plates to tribometer
+    Op->>TR: Set up platter
+    Op->>SW: Activate software
+    Op->>SW: Load plan
+    SW->>TR: Activate tribometer
+    TR-->>SW: Run reciprocating tests (execution loop, see D5)
+    Op->>SW: Software stop
+    SW-->>Op: Save avg files
+    Op->>TR: Remove plates and platters
+    Op->>SW: Close software
+    Op->>TR: Pull equipment out

out/report/diagrams/D5.mermaid

@@ -0,0 +1,8 @@
+flowchart TD
+    L[Load all counterfaces] --> P{5 plates}
+    P -->|for each plate| PR{6 probes in parallel}
+    PR -->|for each probe| C{4 coupons on coupon square}
+    C -->|for each coupon| T[Draw track]
+    T --> RC[Rotate counterface - fresh ball per track]
+    RC -->|3 tracks per coupon| T
+    T --> DONE([120 coupons x 3 tracks = 360 tracks per run])

out/report/diagrams/D6.mermaid

@@ -0,0 +1,13 @@
+erDiagram
+    DEPOSITION_RUN ||--|| SIMTRA_PROFILE : produces
+    DEPOSITION_RUN ||--|| BATCH : creates
+    BATCH ||--|{ WAFER : contains
+    WAFER ||--|{ COUPON : contains
+    COUPON ||--|| XRF_MAP : has
+    COUPON ||--|| NANOINDENTATION : has
+    COUPON ||--|| AFM : has
+    COUPON ||--|| PROFILOMETRY : has
+    COUPON ||--o{ TRACK : "friction coupons only"
+    TRACK ||--|{ CYCLE : has
+    TRACK ||--|{ LOOP_POINT : has
+    TRACK ||--|| WEAR : has

out/report/process_flow.md

@@ -0,0 +1,127 @@
+# Process Flow - Simulated Tribology Laboratory
+
+Generated by make_process_flow.py (task 02) from out/config/lab_config.yaml.
+Standalone diagram sources: ./diagrams/D1..D6.mermaid.
+
+## D1 - Coupon Assembly
+
+Substrate preparation, adhesion layer, and gradient Pt-Au sputter deposition.
+
+```mermaid
+flowchart LR
+    A([Ti-6Al-4V Base 10x10x3 mm]) --> B[Cleaning]
+    B --> B1([Rinsed in deionized water])
+    B1 --> B2([Sonicated in cleaning solution])
+    B --> C[Smearing Adhesive]
+    C --> C1([Cr adhesive coating])
+    C --> D[Vapor Deposition System<br/>Kurt J. Lesker PVD 200]
+    D --> D1([PtAu sputter coating, gradient])
+    D --> E([Test Coupon])
+```
+
+## D2 - Characterization and Batch Assembly
+
+Per-coupon simulation and thickness characterization, then assembly into wafers (x49 coupons) and batches (x3 wafers), deposited per the parameter table below.
+
+```mermaid
+flowchart LR
+    TC([Test Coupon]) --> S[SIMTRA simulation<br/>composition / atom energies]
+    S --> OP[Optical profilometry<br/>film thickness 0.3-1.1 um]
+    OP --> TW[Assemble Test Wafer<br/>x49 coupons]
+    TW --> TB[Assemble Test Batch<br/>x3 wafers]
+    TB --> B721[B721<br/>Pt 150 W / Au 50 W]
+    TB --> B722[B722<br/>Pt 100 W / Au 100 W]
+    TB --> B723[B723<br/>Pt 150 W / Au 50 W]
+    TB --> B724[B724<br/>Pt 50 W / Au 150 W]
+```
+
+Deposition matrix:
+
+| Batch | Pt:Au gun tilt | Pt power W | Au power W | Pt discharge V | Au discharge V |
+|---|---|---|---|---|---|
+| B721 | 20:0 deg | 150 | 50 | 432 | 311 |
+| B722 | 20:20 deg | 100 | 100 | 399 | 352 |
+| B723 | 20:20 deg | 150 | 50 | 430 | 311 |
+| B724 | 0:20 deg | 50 | 150 | 376 | 340 |
+
+## D3 - Testing Tree
+
+The four characterization/testing paths every coupon can take.
+
+```mermaid
+flowchart TD
+    TC([Test Coupon]) --> F[Friction<br/>RAPID 6-probe tribometer]
+    F --> FA([Lab Air -> COF dataset mu_normal])
+    F --> FN([Dry N2 -> COF dataset mu_dry_nit])
+    TC --> NI[Nanoindentation<br/>Bruker TI980]
+    NI --> NIR([Hardness, reduced modulus<br/>25 indents per coupon])
+    TC --> AF[AFM]
+    AF --> AFR([Topography -> Ra / Rq])
+    TC --> XR[micro-XRF<br/>Bruker M4 Tornado]
+    XR --> XRR([Pt/Au wt% map, 20x20 grid])
+```
+
+## D4 - Tribometer Session Sequence
+
+Operator workflow for one RAPID tribometer session, from test request to teardown.
+
+```mermaid
+sequenceDiagram
+    actor Op as Operator
+    participant Prep as Sample Prep
+    participant Plan as Excel Test Plan
+    participant SW as Control Software
+    participant TR as RAPID Tribometer
+
+    Op->>Prep: Request test
+    Op->>Prep: Get samples
+    Op->>Prep: Load samples onto plates
+    Op->>Prep: Load ball holders (3 counterfaces per holder)
+    Op->>Plan: Create test plan<br/>(Sample Plate, Plate Location, Sample ID, Save Location,<br/>Folder Name, X/Y/Z Offset, Load, Iterations)
+    Op->>TR: Transfer plates to tribometer
+    Op->>TR: Set up platter
+    Op->>SW: Activate software
+    Op->>SW: Load plan
+    SW->>TR: Activate tribometer
+    TR-->>SW: Run reciprocating tests (execution loop, see D5)
+    Op->>SW: Software stop
+    SW-->>Op: Save avg files
+    Op->>TR: Remove plates and platters
+    Op->>SW: Close software
+    Op->>TR: Pull equipment out
+```
+
+## D5 - Execution Loop
+
+Nested iteration executed by the tribometer within one run.
+
+```mermaid
+flowchart TD
+    L[Load all counterfaces] --> P{5 plates}
+    P -->|for each plate| PR{6 probes in parallel}
+    PR -->|for each probe| C{4 coupons on coupon square}
+    C -->|for each coupon| T[Draw track]
+    T --> RC[Rotate counterface - fresh ball per track]
+    RC -->|3 tracks per coupon| T
+    T --> DONE([120 coupons x 3 tracks = 360 tracks per run])
+```
+
+## D6 - Data Hierarchy
+
+Entity containment and per-entity measurement datasets (ERD-style).
+
+```mermaid
+erDiagram
+    DEPOSITION_RUN ||--|| SIMTRA_PROFILE : produces
+    DEPOSITION_RUN ||--|| BATCH : creates
+    BATCH ||--|{ WAFER : contains
+    WAFER ||--|{ COUPON : contains
+    COUPON ||--|| XRF_MAP : has
+    COUPON ||--|| NANOINDENTATION : has
+    COUPON ||--|| AFM : has
+    COUPON ||--|| PROFILOMETRY : has
+    COUPON ||--o{ TRACK : "friction coupons only"
+    TRACK ||--|{ CYCLE : has
+    TRACK ||--|{ LOOP_POINT : has
+    TRACK ||--|| WEAR : has
+```

out/report/tables/t1_storage_footprint.csv

@@ -0,0 +1,6 @@
+format,measured,vs CSV,archival (compressed),projected @600 GB,projected @1.2 TB,projected @6 TB
+CSV,149 MB,1.00x,53 MB,600 GB,1.20 TB,6.00 TB
+JSON-LD,311 MB,2.09x,57 MB,1.26 TB,2.51 TB,13 TB
+SQLite,123 MB,0.83x,55 MB,498 GB,996 GB,4.98 TB
+PostgreSQL,397 MB,2.67x,53 MB,1.60 TB,3.21 TB,16 TB
+RDF triplestore,1.74 GB,11.70x,126 MB,7.02 TB,14 TB,70 TB

out/report/tables/t2_measured_medians.csv

@@ -0,0 +1,36 @@
+format,query,wall time,peak RSS,CPU util,read
+CSV,q1,0.06 s,4 MB,0%,0.0 MB
+CSV,q2,0.17 s,20 MB,16%,2.5 MB
+CSV,q3,1.15 s,21 MB,24%,31.8 MB
+CSV,q4,0.71 s,21 MB,22%,16.8 MB
+CSV,q5,1.09 s,21 MB,23%,31.8 MB
+CSV,q6,0.28 s,20 MB,17%,1.2 MB
+CSV,q7,1.27 s,21 MB,23%,32.5 MB
+JSON-LD,q1,0.11 s,22 MB,10%,1.4 MB
+JSON-LD,q2,1.48 s,24 MB,24%,53.1 MB
+JSON-LD,q3,9.94 s,24 MB,25%,297.6 MB
+JSON-LD,q4,9.92 s,24 MB,25%,297.8 MB
+JSON-LD,q5,9.96 s,24 MB,25%,297.6 MB
+JSON-LD,q6,9.79 s,24 MB,24%,297.8 MB
+JSON-LD,q7,10 s,24 MB,24%,297.9 MB
+SQLite,q1,0.06 s,4 MB,0%,0.0 MB
+SQLite,q2,0.06 s,4 MB,0%,0.0 MB
+SQLite,q3,0.11 s,4 MB,0%,0.0 MB
+SQLite,q4,0.06 s,4 MB,0%,0.0 MB
+SQLite,q5,0.77 s,25 MB,23%,117.9 MB
+SQLite,q6,0.06 s,4 MB,0%,0.0 MB
+SQLite,q7,0.71 s,25 MB,23%,93.0 MB
+PostgreSQL,q1,0.22 s,42 MB,18%,4.1 MB
+PostgreSQL,q2,0.22 s,42 MB,20%,4.1 MB
+PostgreSQL,q3,0.22 s,42 MB,18%,4.1 MB
+PostgreSQL,q4,0.22 s,42 MB,18%,4.1 MB
+PostgreSQL,q5,0.44 s,42 MB,10%,4.1 MB
+PostgreSQL,q6,0.22 s,42 MB,19%,4.1 MB
+PostgreSQL,q7,0.33 s,42 MB,12%,4.1 MB
+RDF triplestore,q1,0.11 s,32 MB,17%,3.6 MB
+RDF triplestore,q2,2.81 s,85 MB,24%,375.6 MB
+RDF triplestore,q3,46 s,264 MB,25%,5445.1 MB
+RDF triplestore,q4,20 s,168 MB,25%,2694.6 MB
+RDF triplestore,q5,60 s,264 MB,25%,5390.5 MB
+RDF triplestore,q6,0.17 s,46 MB,16%,15.9 MB
+RDF triplestore,q7,41 s,264 MB,25%,5380.5 MB