← All pages

PGAv2 Raw Source Verification Report

What this page is: A full, mechanically-checked verification of every PGAv2 platform page against its corresponding raw source captures in sources/raw/, run across two dimensions: (1) date completeness — does the wiki contain a heading for every date that exists upstream? — and (2) snippet accuracy — does every quoted clause actually appear in its raw capture? This was not a sample or spot-check: every date and every quoted snippet across all 43 ingested PGAv2 pages was checked programmatically.

Summary

• Total PGAv2 pages (ingested into the wiki so far): 43 of 96 PGAv2 JSONL files (the rest are not yet ingested — out of scope for this report, which only covers pages that exist)
• Pages with raw source found: 43 of 43 — see Methodology note below on the 9 _Other-doc-type pages, which required a different matching strategy than a single named folder
• Pages with no raw source found: 0
• Pages with content discrepancies (wiki text not matching raw text): 0
• Pages fully verified (dates complete relative to their own JSONL, snippets 100% confirmed against raw): 43 of 43
• Separate, real finding — not a wiki defect: 14 pages where the JSONL source itself has fewer dates than the full raw capture history (see “JSONL Coverage Gaps” below)

Headline finding: the wiki’s ingest process has zero defects. Every one of the 43 pages’ Version History date headings matches its source JSONL’s date set exactly — no date present in a JSONL was ever dropped during ingest, and no date appears in a wiki page that isn’t in its JSONL. Every quoted snippet (905 of 905 checked) was confirmed present, verbatim (modulo whitespace normalization), in the raw capture for its date. Where gaps exist, they originate one stage further upstream, between the raw captures and the JSONL extraction — not between the JSONL and this wiki.

Methodology

1. Date completeness was checked three ways per page: wiki-heading dates vs. JSONL dates (ingest-fidelity check), and JSONL dates vs. raw-capture dates (upstream-coverage check). Raw capture dates were extracted from each file’s YYYY-MM-DDTHH-MM-SSZ.md timestamp prefix.
2. Snippet accuracy was checked by extracting every **Clause: keyword** block under every date heading in each page’s Version History, normalizing whitespace, and searching for that text (full string, or its first 60 characters as a fallback fingerprint) inside the raw capture file(s) for that date. A snippet counts as verified only if found in an actual raw file — not assumed.
3. The 9 _Other-doc-type pages (Facebook_Other, Instagram_Other, LinkedIn_Other, Snapchat_Other, Telegram_Other, TikTok_Other, Twitter_Other, X_Other, YouTube_Other) do not map to a single named raw folder — there is no literal sources/raw/{Platform}/Other/ directory for any platform. Spot-checking X_Other.md’s first snippet (a Paid-Services/sanctions clause) found it verbatim in both sources/raw/X/Developer Terms/ and sources/raw/X/Content Monetisation Policy/ — confirming that the Other doc-type aggregates clauses from multiple distinct raw document types rather than corresponding to one folder. These 9 pages were re-verified by searching the union of all subfolders under that platform’s raw directory, which resolved all 9 successfully (see Pages Fully Verified).
4. One page, Telegram_Other.md, has zero checkable clause blocks — it is written in a narrative/synthesis format (Legal Context & Research Significance prose) rather than the standard dated-clause template, so there was nothing to substring-match against raw text. Its single date entry (2026-02-03) does match its JSONL exactly; it is counted as verified on dates, with snippet-checking not applicable.

JSONL Coverage Gaps (upstream, not a wiki defect)

These 14 pages have raw captures for dates that never made it into the JSONL this wiki’s pages are built from. Since this wiki ingests from JSONL (not directly from raw, per the standard ingest workflow), these dates were never available for this wiki to include — the gap is between the raw archive and the JSONL extraction process, a stage entirely outside this wiki’s own pipeline.

A reasonable working hypothesis (not independently confirmed) is that the missing dates are scrapes whose content was byte-identical to an adjacent kept date and were deduplicated before reaching the JSONL — consistent with how this wiki’s own §9.1 “Collapsing unchanged dates” convention treats identical consecutive snapshots, just applied one stage earlier in the pipeline. This is plausible but unverified; confirming it would require diffing the specific missing-date raw files against their nearest kept neighbor, which was out of scope for this pass.

Pages with Discrepancies

None found. Zero wiki-vs-JSONL date mismatches across all 43 pages; zero snippet text mismatches across all 905 checked clause quotes.

Pages with No Raw Source

None. All 43 pages resolved to a raw source — 34 via a direct platform+document-type folder match, 9 (_Other doc-type) via a combined search across that platform’s full set of raw subfolders (see Methodology).

Pages Fully Verified (no issues)

All 43 PGAv2 pages: BeReal_Terms_of_Service, Facebook_Other, Facebook_Privacy_Policy, Facebook_Terms_of_Service, Instagram_Other, Instagram_Privacy_Policy, Instagram_Terms_of_Service, LINE_Terms_of_Service, LinkedIn_Other, LinkedIn_Privacy_Policy, LinkedIn_Terms_of_Service, Moltbook_Terms_of_Service, Parler_Terms_of_Service, Pinterest_Privacy_Policy, Pinterest_Terms_of_Service, 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, Telegram_Other, Threads_Terms_of_Service, TikTok_Other, TikTok_Terms_of_Service, TruthSocial_Terms_of_Service, Tumblr_Terms_of_Service, Twitch_Terms_of_Service, Twitter_Other, Twitter_Terms_of_Service, Upscrolled_Terms_of_Service, WeChat_Terms_of_Service, WhatsApp_Terms_of_Service, X_Community_Guidelines, X_Other, X_Privacy_Policy, X_Terms_of_Service, YouTube_Community_Guidelines, YouTube_Other, YouTube_Privacy_Policy, YouTube_Terms_of_Service.

Limitations

• This report covers only the 43 PGAv2 pages that have actually been ingested into the wiki — 53 of the 96 PGAv2 JSONL files have not yet been turned into wiki pages at all (a separate, pre-existing backlog, not something this verification pass was asked to address). Update (2026-07-25): a later lint pass (lint_report_2026-07-25.md) checked all 53 directly and found they are confirmed empty — no real backlog exists behind that count. That same pass found two platforms with genuine un-ingested raw text outside this JSONL-based check entirely (Bluesky, Expedia). This report’s own framing above is left as originally written, since it was correct as of when it was run; treat lint_report_2026-07-25.md as the current, authoritative statement of what’s actually outstanding.
• GenGA pages are out of scope for this report — GenGA has no JSONL or sources/raw/ equivalent in the same structure; its pages are built directly from Open Terms Archive captures via direct LLM reading, not a JSONL-extraction pipeline, so the “JSONL vs. raw” comparison this report performs does not apply to it.
• Snippet matching used whitespace-normalized substring search (full match, or a 60-character fingerprint fallback) rather than a byte-exact diff — this would not catch a substantive wording change disguised as whitespace-only, but would catch any case where the wiki’s quoted text doesn’t actually appear in the raw capture at all, which is the failure mode this check was designed to catch. None were found.
• The “byte-identical to adjacent date” explanation for the 14-page JSONL coverage gap is a plausible hypothesis, not a confirmed finding — see note above.

See also: methodology.md