← All pages

Wiki Health Check — 2026-06-21

A structured lint pass per CLAUDE.md §5.3, covering all 99 platform pages and 27 concept pages (at the time of this check). Performed programmatically wherever a check was mechanically verifiable, to avoid the risk of a manual spot-check missing something at this scale.

1. Orphan check — PASS

Every file in pages/platforms/ and pages/concepts/ is linked from pages/index.md. Verified by cross-checking every filename in both directories against index.md’s content. Zero orphans found.

2. Threshold check — 2 GAPS FOUND AND FIXED

CLAUDE.md §4.3 requires a concept page once a keyword appears in 3+ platform pages. A full keyword-frequency scan across every page’s Flagged Keywords section found two keywords well past this threshold with no concept page:

• input license — 21 platform pages, no concept page existed. Fixed: created input_license.md, linked from all 21 pages.
• output ownership — 10 platform pages, no concept page existed. Fixed: created output_ownership.md, linked from all 10 pages.

Two other apparent gaps turned out to be non-issues on inspection: prepare derivative (4 pages) is already explicitly handled as a named variant inside derivative_works.md and create_derivative.md; risk_score, keyword, and genai-eu are boilerplate methodology-note text, not actual flagged-keyword tags, and were excluded.

3. Missing cross-reference check — 2 BROKEN LINKS FOUND AND FIXED

A full link-resolution scan across every .md file in pages/ found exactly 2 broken links: Le_Chat_Commercial_Terms.md and Microsoft_Copilot_Terms_of_Service.md both pre-emptively linked to a then-nonexistent ../concepts/input_license.md, each marked “(if threshold reached)”. Both resolved automatically once input_license.md was created in the threshold-check fix above; the now-redundant “(if threshold reached)” annotations were removed. Zero broken links remain (re-verified after the fix).

A further script-driven pass added the same input license/output ownership concept links to all 21 + 10 platform pages using each keyword, matching this wiki’s existing cross-linking style (bolded bullet headers and Version History **Clause:** headers only — the plain **Flagged keywords:** summary lists were deliberately left as bare backticks, consistent with how every other keyword in this wiki is already handled).

4. Staleness check — PASS

Cross-checked every row in index.md’s two summary tables (99 rows) against the actual latest ### {date} heading in the corresponding platform page’s Version History. 0 mismatches — every “Latest Date” in the index is exactly the most recent dated entry in the source page.

5. Contradiction check — 1 DRIFT FOUND AND FIXED, otherwise PASS

Checked every concept page’s stated running count (“confirmed across N platform pages”) against the actual row count in its own “Platforms Using This Clause” table, and cross-checked the synthesis page genga_vs_pgav2_comparison.md’s cited counts (13/9/14/2) against current concept-page tables.

• Found: input_license.md’s own Notes & Trends section and its index.md summary line both said “19 platform pages,” but the table itself (built from a separate, more precise grep pass) actually has 21 rows. Fixed in both locations.
• All other counts checked (feedback=18, ai_disclosure=9, competing_model_ban=13, gdpr=14, output_ownership=10) matched exactly. The synthesis page’s cross-dataset counts (13 for competing-model-ban, 9 for ai-disclosure, 14 for GDPR, 2 for EU AI Act citations) were independently re-verified via grep against current platform pages and confirmed still accurate — no drift since that page was written.

6. Formatting check — PASS

Scanned all of pages/ for literal tab characters, trailing whitespace, and 3+ consecutive blank lines. Zero instances of any of the three found — formatting is already consistent across all 126 files.

7. Naming inconsistency — 1 MINOR ISSUE FOUND, NOT YET FIXED

Upscrolled_Terms_of_Service.md’s filename and **Source:** field use “Upscrolled” (matching the canonical sources/jsonl/Upscrolled_*.jsonl filenames), but the page’s own H1 title and both index.md references use “UpScrolled” (camelCase). This is cosmetic, not a data error — the link itself correctly resolves to the right file — but is a naming inconsistency worth normalizing to match the source filename casing. Flagged here as a small open item, not fixed in this pass since it touches the page title and is lower priority than the threshold/cross-reference gaps above.

Summary

The wiki was already in strong structural health going into this pass — the two real gaps (input_license.md, output_ownership.md) were genuine, traceable to the dataset’s keyword vocabulary having grown ahead of concept-page creation during the GenGA ingest, not to any systemic process failure. All fixes are additive and preserve every existing piece of legal-text quoting, per CLAUDE.md’s “never paraphrase legal text” rule.