[percosys]

Schema part of #892. The data follows in a separate PR on top of this one.

What

An optional mechanical array on filament.json (schemas/mechanical_schema.json). One entry per published value: property, print direction, value, ± spread, unit, standard, dry or conditioned, test-bar details (specimen), and the same source fields as sustainability (source_type, source_url, source_quote, retrieved, notes).

It follows #768:

• Schema: linked from the filament schema with a cross-file $ref. ofd-validator resolves it as is (tested).
• Validation (data_quality.py): reuses the sustainability source checks. Errors: missing source fields, value or ± not in the quote, retrieved in the future, a unit that doesn't fit the property. Warnings: a print direction on moulded bars, a repeated entry (same property, direction, condition, standard, unit and specimen).
• Review and merge: the Sustainability sources gate also checks mechanical entries against their source (the quote on the page, the value in the quote), and a PR that touches them never auto-merges. The AI review never suggests mechanical values, and its mechanical findings are never dropped. The gate fails closed on this PR because it changes rule files (data_quality.py, sustainability.py), so a maintainer must review it.
• Number in quote: in MPa, psi, ksi and J/m it also matches grouped thousands (7250 ↔ 7,250), as sheets in psi write them. In other units a comma stays a decimal comma. A value no longer matches part of a longer number (5 in 5.85, 7 in 7,250). Sustainability quotes never group.
• Exports: a mechanical JSON column in SQLite and CSV. The API carries the field as is.
• style_data: key order from the schema; tracking parameters stripped from source_url; specimen cleaned like the other fields.
• WebUI: a table on the filament page (XY / XZ / Z side by side, each value linked to its source), an editor in the filament form (reuses SourcedEntryFields), the schema endpoint, the same checks client-side, and tests. A new entry left empty is dropped on save.
• Docs: a "🔩 Mechanical data" section in the manual.

Notes

• Filament only. A data sheet covers a product line, so variants don't override it.
• Units stay as published (psi stays psi): copy, never compute.
• Changes from the issue: direction gains XZ (some sheets test on-edge bars) and uses unknown instead of "".
• Two small WebUI fixes this needed, which also touch sustainability:
  • checkSourcedEntry checked every value field except value_kg_co2e as a percentage. It now checks only pct, recycled_pct and reduction_pct, like the Python side.
  • A required dropdown used to pick its first option silently (a new carbon footprint got method: "epd"). It now starts on "Choose…" and shows an inline "… is required" error until a value is picked.

Testing

• pytest: 547 passed. The 2 failures in test_search_index_exporter.py also fail on main.
• ruff check, ruff format --check: clean.
• ofd validate: passes with no errors. style_data --dry-run: 0 files changed.
• WebUI: vitest 1071 passed, npm run check 0 errors, npm run build OK.

🤖 Generated with Claude Code

[coffandro]

Thanks for this. The mechanical schema and the editor look solid. Below are the issues I found. Inline comments cover the changed lines. Two concerns are about code this PR doesn't touch, so they're here:

AI review guard skips mechanical. In ofd/review/suggest.py (~L138), the "never suggested" filter blocks pointers containing sustainability or container_details, but not mechanical. A review model could emit an enrichment like /mechanical/0/value with a guessed number, and it would become a PR suggestion. That breaks the "copy, never compute" rule this schema enforces. ofd/review/triage.py (~L73) and ofd/review/sustainability.py (~L448) have the same sustainability-only checks.

No source gate or auto-merge exclusion for mechanical-only PRs. The "Sustainability sources" gate (ofd/review/sustainability.py ~L60) and auto-merge eligibility (eligibility.py ~L285) only look at sustainability paths. A PR that only adds mechanical entries with a wrong source_url, where the quote check still passes, gets no human source review and can be auto-merged. The PR description mentions this gap. I'd close it here or in a follow-up before contributors start using the field.

I think the two above and the two number-matching comments on _number_pattern should be fixed before merge. The rest are smaller.

🤖 Generated with Claude Code

[github-actions]

Automated review of 1511610 · how this works

• 🌱 Sustainability: 2 problem(s) overridden by @coffandro.

Maintainers: ✅ LGTM & merge · _{workflow run}

[percosys]

Thanks for the review. All points are fixed in 1511610.

• AI review guard: suggest.py now rejects any /mechanical pointer ("mechanical data is never suggested"). triage.py and model_problems in sustainability.py now use one is_sourced_pointer() that covers sustainability, container_details and mechanical, so mechanical findings are never dropped.
• Source gate and auto-merge: the "Sustainability sources" gate now also checks each mechanical entry against its source (the quote on the page, the value in the quote). A PR that adds, changes or removes mechanical data counts as touched, so E15 keeps it from auto-merging. schemas/mechanical_schema.json is now a rule file. The "not covered" note is gone from the description.
• Number matching: see the two threads on _number_pattern.

The bot's "Sustainability blocked" on this PR is expected: the gate fails closed when data_quality.py or sustainability.py change, until a maintainer reviews them.

Tests: pytest 547 passed (the 2 failures in test_search_index_exporter.py also fail on main). ruff is clean. ofd validate passes with no errors, and style_data --dry-run changes 0 files. WebUI: vitest 1071 passed, npm run check 0 errors, npm run build OK.

🤖 Addressed by Claude Code

[coffandro]

LGTM! Thanks for the contribution, it'll be up on SP and the dataset at about 00:00 Danish time.
You can make a PR for the data additions aswell if you'd like :D