← Back to /admin

2026-09-18-wisdom-rank9plus.md

Audit — Wisdom Rank-9+ Insights — 2026-09-18

Target. 25 insights with impact_rank ≥ 9 sampled from wisdom/insights.json (via audits/inputs-2026-09-18/wisdom-sample.json).

Prior audit. audits/2026-04-24-wisdom-rank9plus.md — 14/20 passed on that run. Six mispillared/rank-inflated, all attributable to substring-matching classifier drift.

This run's verdict. Verbatim accuracy holds. Pillar drift is much worse than the prior sample suggested — 13/25 mispillared, matching the pre-existing 48% finding from audits/2026-09-18-pillar-convergence-plan.md.

Summary

Axis	Pass	Note	Fail	Notes
Text verbatim	25	—	0	Zero fabrications. Every insight is a legitimate paragraph-level extract.
Rank ≥ 9 deserved	15	3 borderline	7 inflated	Same over-rank pattern S14 flagged: AML transcripts, framing paragraphs, duplicates.
Pillar correct	9	3 arguable	13 wrong	Overwhelmingly law-governance content misfiled as bitcoin-protocol.
Shape (insight vs boilerplate)	24	—	1	INS-0050 (Trotskyite/Big-Brother polemic) is emphatic description, not a claim.

Top-5 mistagged insights

1. INS-0003 (rank 10, bitcoin-protocol) — Verbatim quote of De Filippi 2016 AML-pseudonymity paragraph inside an AML/scaling paper. Not a foundational Wright thesis. → rank 7-8, pillar law-governance. Same insight S14 flagged; still ranked 10.

2. INS-0011 (rank 9, bitcoin-protocol) — Textbook definition of hearsay + email example. Zero protocol content. → rank 6, pillar law-governance.

3. INS-0041 (rank 9, bitcoin-protocol) — Passing-off / database-rights / MIT-License distinction. Pure law-governance. S14 flagged this exact drift; still wrong. → pillar law-governance.

4. INS-0050 (rank 9, bitcoin-protocol) — Trotskyite/Big-Brother polemic aimed at BTC-Core-supporting lawyers. Emphatic description, not a claim. → rank 6-7, pillar law-governance, shape boilerplate.

5. INS-0036 (rank 9, bitcoin-protocol) — Tominaga Nakamoto's history-instability thesis and Wright's naming choice. Tagged bitcoin-protocol from surface term 'protocol' (used metaphorically for language). → pillar identity-history.

Patterns

1. Law-governance drift is systemic on this sample. 10 of the 13 pillar-wrong insights should be law-governance. The substring matcher catches keywords like blockchain, nodes, protocol in what is otherwise pure legal-doctrine writing. Consistent with the 48% blog-level finding in pillar-convergence-plan.md.

2. AML/regulatory transcripts continue to over-rank. INS-0003 and INS-0057 both quote external-authority AML paragraphs (De Filippi, Laroiya) but inherit rank ≥ 9 because they are long, formal, and prominently placed. S14 flagged the same pattern; unchanged.

3. Duplicate/near-duplicate insights. INS-0010 vs INS-0028 (fair-exchange framing repeated across two posts); INS-0007 vs INS-0008 (electronic-contracts paragraph recycled). Two rank-9s for one thesis = wisdom-dedup opportunity.

4. identity-history and economics under-attributed. INS-0036 belongs to identity-history; INS-0013, INS-0012, INS-0014 to economics — all are being pulled into bitcoin-protocol by weak surface matches.

5. Verbatim discipline held. Session 7 cleanup and downstream extractors are still producing faithful passages. The failure mode remains classification, not fabrication.

Recommendations

• Adopt the canonical 10-pillar list from audits/2026-09-18-pillar-convergence-plan.md; re-score the whole wisdom/insights.json with word-boundary matching plus mathematics/enterprise support. The current pipeline is silently dropping two pillars.
• Down-rank block-quoted regulatory-transcript paragraphs: rank-10 should be reserved for Wright's one-line thesis statements ("It is not cryptography, it is economic incentives"; "capitalist money designed to follow Mises and Hayek"; "Bitcoin is not democracy"), not any long-quotable citation.
• De-dup at extraction time (near-duplicate paragraph across sources = one insight with also_at list, not two).
• Text verbatim gate is working; no cleanup needed on that axis.

Re-run quarterly.