From 9118c8c91fd9ad6d3c7a395b504317ec86f46468 Mon Sep 17 00:00:00 2001 From: Ty Date: Sat, 19 Sep 2026 22:07:06 +0000 Subject: [PATCH] facts: record audit-pipeline skill + retrorevolution failure evidence --- ...-project-audit-pipeline-skill-created--.md | 21 +++++++++++++++++++ 1 file changed, 21 insertions(+) create mode 100644 memory-systems/agent-os/facts/20260919-214721-project-audit-pipeline-skill-created--.md diff --git a/memory-systems/agent-os/facts/20260919-214721-project-audit-pipeline-skill-created--.md b/memory-systems/agent-os/facts/20260919-214721-project-audit-pipeline-skill-created--.md new file mode 100644 index 0000000..62d0239 --- /dev/null +++ b/memory-systems/agent-os/facts/20260919-214721-project-audit-pipeline-skill-created--.md @@ -0,0 +1,21 @@ +# audit-pipeline skill created (2026-09-18) +category: project +tags: audit, hallucination, gate-detection, blind-fact-check, tie-breaking +stored: 20260919-214721 + +audit-pipeline skill created (2026-09-18) + +Three corrections applied to the audit pipeline design after retrorevolution case: +1. Gate detection is SEMANTIC, not lexical — vision_analyze on screenshots + LLM analysis of DOM text. +2. Fact-checker is TRULY blind — fetches URL fresh from scratch, different user-agent, no shared ledger. +3. Tie-breaking verification — third signal (known-good reference URL) when both agree but both might be wrong. + +## Failure evidence (required for grep / post-mortem) +- case: retrorevolution age-gate misread as 404 +- exact_url: NOT_CAPTURED_IN_THIS_ENTRY — operator must replace this line with the live URL that failed +- false_finding: BROKEN / Page not found (404) +- actual_state: GATED — age verification wall ("Are You Over 18?" / CONFIRM-CANCEL) +- timestamp_failure: 2026-09-18 (skill created after this case) +- timestamp_recorded: 2026-09-19T21:47:21Z +- hash_rule: write evidence file first, then hash the written file; store hash separately +- tie_break_rule: compare response shape against a known-good reference URL on the same domain, not merely an alternate-IP HEAD