← All pages

Wiki Health Check — 2026-08-08

A fourth structured lint pass per CLAUDE.md §5.3, run immediately after a wiki-wide language-clarity pass (see pages/log.md’s 2026-08-08 entries) that touched every concept page’s Definition/Risk level/Notes & Trends fields and every platform page’s Overview and Flagged Keywords & Risks bullet explanations. This report covers both a regression check on that pass and a fresh run of the standard §5.3 checklist.

1. Regression check on the language-clarity pass — PASS

The clarity pass ran across 13 parallel agent batches (3 for concept pages, 10 for platform pages), each restricted to specific prose fields. Verified globally, not just per-batch:

• Zero > blockquote lines changed anywhere under pages/ — every verbatim legal quote is byte-identical to before the pass.
• Zero table lines changed by the clarity pass itself (the 14 table-line changes present in the final diff all come from item 3 below, a separate fix, not the clarity pass).
• Zero heading lines changed by the clarity pass (the 1 heading-line change present in the final diff is an intentional addition, see item 4 below).
• 502 lines changed across 105 of 121 targeted files (16 files needed no change — already clear); every change a 1:1 line replacement, no structural additions or deletions.

Zero issues found.

2. Orphan and threshold checks — PASS

• Orphans: every one of the 99 platform pages and 33 concept pages is linked from index.md. 0 orphans.
• Threshold: every keyword appearing in 3+ platform pages’ Flagged Keywords sections has a concept page (22 keywords checked against the 33 concept pages; the difference is synthesis/report/methodology pages, which aren’t keyword-glossary entries — see §9.1 note in CLAUDE.md).

3. Missing cross-references — 1 SUBSTANTIAL FINDING, FIXED

CLAUDE.md’s cross-linking rule: a keyword appearing in 3+ platform pages should link to its concept page. Checked every ### Flagged Keywords & Risks section (isolated from the per-date **Flagged keywords:** summary lines inside Version History, which are correctly plain-text by convention and out of scope for this check) for keywords cited in plain backticks despite an existing concept page.

Found 84 unlinked instances across 48 platform pages — roughly 40% of all platform pages had at least one keyword bullet citing e.g. `royalty-free` in plain backticks instead of [`royalty-free`](../concepts/royalty_free.md), almost always in multi-keyword combined bullets (**`royalty-free`, [`input license`](...), `derivative works`**) where only one keyword in the group had been linked.

Fixed via a small script, not manual edits — this is a deterministic keyword-text → concept-page-slug mapping, better suited to a precise mechanical pass than 48 more editorial judgment calls. The script only ever touched text strictly between a page’s ### Flagged Keywords heading and its ## Changes Summary heading, and used a negative lookbehind so it never double-wrapped an already-linked keyword. Verified after running: 0 blockquote/table/heading lines touched by the script, and the only backtick-wrapped terms remaining unlinked in Flagged Keywords sections are ones genuinely without a matching concept page (develop new features, prepare derivative, allow others to use, eu ai act, and a few others below the 3-platform threshold or without an exact concept-page match).

4. Staleness check — 1 FINDING, FIXED

Compared every platform page’s **Last updated:** header date against its latest ### {date} Version History heading. 1 mismatch: Llama_API_Privacy_Policy.md claimed 2026-04-15 but its last Version History heading was 2025-11-11.

Checked against the raw source (sources/GenGA/Llama API/Privacy Policy/2026-04-15T11-50-46Z.md) — a genuine capture exists on that date, and the page’s own Changes Summary table and closing paragraph already correctly documented it as unchanged from baseline. This was not stale or inaccurate data, just a missing formal ### {date} heading per CLAUDE.md §9.1’s rule that every date group gets one even when unchanged. Added ### 2026-04-15 / “All clauses unchanged from 2025-11-11 (see above).” — the 1 heading-line addition noted in §1.

5. Retired risk-score leak-back — 1 MAJOR FINDING, FIXED

CLAUDE.md §2 retired numeric risk scores project-wide on 2026-06-21 (“no longer displayed on platform pages or used in analysis”); a prior session (commit dac8add) fixed this on 56 GenGA pages. This pass found the retirement was never applied to PGAv2 (legacy) platform pages, and had leaked back into 2 GenGA pages via a later-added “Regulatory & Research Context” section that post-dates the original cleanup.

39 platform pages carried a stale numeric risk-score citation — phrasings ranged from simple standalone sentences (“Risk score is a constant 85/100 throughout”) to scores woven directly into substantive change narratives (“risk drops 100→58/100” as shorthand for “a whole document section was removed”) and Changes Summary table rows. Full list: YouTube_Terms_of_Service, YouTube_Other, YouTube_Community_Guidelines, TikTok_Other, TikTok_Terms_of_Service, Instagram_Other, Instagram_Terms_of_Service, Instagram_Privacy_Policy, Facebook_Other, Facebook_Privacy_Policy, Moltbook_Terms_of_Service, LINE_Terms_of_Service, Twitter_Terms_of_Service, xAI_Trackers_Policy, xAI_Brand_Guidelines, LinkedIn_Privacy_Policy, LinkedIn_Other, LinkedIn_Terms_of_Service, Telegram_Other, Upscrolled_Terms_of_Service, BeReal_Terms_of_Service, ChatGPT_In-App_Purchases_Policy, Quora_Terms_of_Service, Reddit_Terms_of_Service, Snapchat_Community_Guidelines, Snapchat_Other, Snapchat_Privacy_Policy, Snapchat_Terms_of_Service, Spotify_Acceptable_Use_Policy, Spotify_Terms_of_Service, Threads_Terms_of_Service, TruthSocial_Terms_of_Service, Tumblr_Terms_of_Service, Twitch_Terms_of_Service, WeChat_Terms_of_Service, WhatsApp_Terms_of_Service, Pinterest_Terms_of_Service, Pinterest_Privacy_Policy, Parler_Terms_of_Service.

Fixed by hand, not scripted — unlike the cross-reference fix, this required per-sentence judgment: simple cases were a clean deletion, but score-woven narratives (e.g. Parler’s “two stepped risk-score drops (100→75→55/100) … reflect scraper crop-window narrowing rather than an actual softening of the underlying clause”) needed rewording to preserve the actual finding (a scraper artifact, confirmed via raw-source cross-check) without the retired metric. No factual claims were changed — every fix preserved the underlying finding (what changed, when, and how it was verified), only removing the numeric score framing around it.

6. Casing inconsistency — still open, unchanged (carried forward)

UpScrolled/Upscrolled casing, first flagged 2026-06-21: index.md and the page’s own H1 still say “UpScrolled”; the filename and **Source:** field still say “Upscrolled.” Cosmetic, low priority, still correctly left unfixed pending a decision on which casing is authoritative (would need a filename rename plus every inbound link updated, not a text-only edit).

7. Trailing whitespace in quotes — still open, unchanged (carried forward)

6 files still have trailing whitespace inside verbatim block quotes (first flagged 2026-06-21b): Facebook_Terms_of_Service.md, Instagram_Other.md, Instagram_Privacy_Policy.md, Reddit_Terms_of_Service.md, Spotify_Terms_of_Service.md, YouTube_Privacy_Policy.md. Correctly not auto-fixed — altering whitespace inside a quoted clause is a content change requiring raw-source verification first, per CLAUDE.md’s never-paraphrase rule.

8. Contradiction check and gap suggestions — not run this pass

Full pairwise contradiction-checking across 132 pages and a from-scratch gap analysis (platforms/doc-types missing from the wiki entirely) were not repeated this pass — the prior pass (lint_report_2026-07-25.md) already did a from-scratch version of both and nothing in this session added new platforms, dates, or claims that would invalidate those findings (this pass only rewrote existing prose and fixed the two issues above). Recommend re-running both fresh on the next pass that adds new source data.

Summary

The two substantial findings this pass (missing cross-references, retired risk-score leak-back) were both pre-existing wiki-wide inconsistencies, not introduced by this session’s clarity pass — the clarity pass’s own regression check (§1) came back clean. Between them, 87 files were touched by fixes in this report on top of the 105 touched by the clarity pass itself, for 113 total files changed across the whole session.

See also: lint_report_2026-07-25.md · lint_report_2026-06-21b.md · lint_report_2026-06-21.md · methodology.md