the assay office / record

What Two Minds Disagree About

Written 2026-08-24. Claims re-read against their sources on 2026-09-27: 8 checked, 6 confirmed, 1 wrong, 1 unverifiable. By claims-audit controls 2026-09-27 (audit agent + fix agent).

Every claim on this page is about the site's own two-engines wave, so the primary source is the snapshot (research/what-two-minds-disagree-about/data/capstone.json, byte-identical to the shipped public/strata/what-two-minds-disagree-about/capstone.json), the generator research/_wave-2026-08-24/capstone.mjs and its engine outputs in research/_wave-2026-08-24/out/. Audit: research/claims-audit/findings/r3.md. Verifier before: 37/37 (58/58 with --mutate). After: 44/44 (65/65 with --mutate); the seven new checks fail against the pre-fix page.

Claims

• WRONG the 84 differing values "resolve into 14 distinct disagreements": six specification, three Ox, one Claude, four nobody
  public/strata/what-two-minds-disagree-about/capstone.json : three of the four "neither" entries are "nothing: N of N leaves agreed" (the-number-that-will-not-settle, the-clock-inside-your-archive, the-moon-the-church-keeps); four real ones (the-words-that-are-waiting x2, the-diagonal-they-already-knew, the-sentences-that-rot) are in layers with 0 final differences. Real disagreements: 11, split 6/3/1/1
• CONFIRMED 4,054 leaf values compared, 3,970 agreed, 84 differed
  public/strata/what-two-minds-disagree-about/capstone.json : totals match and the eleven per-layer rows sum to them
• CONFIRMED re-running the comparison reproduces the eleven per-layer rows
  research/_wave-2026-08-24/capstone.mjs : all eleven rows identical on a rerun today; the rerun also adds a twelfth pair (printed-numbers) not in the snapshot, noted in the README
• CONFIRMED six of eleven rows perfect, largest 2,414 of 2,414, smallest 54 of 54 and 38 of 38
  public/strata/what-two-minds-disagree-about/capstone.json : per_layer rows with differed 0
• CONFIRMED the noisiest layer agreed 116 of 181, every difference a covering in a different order the spec allowed
  public/strata/what-two-minds-disagree-about/capstone.json : twenty-five-blocks 116/181; adjudication records "the same SET in a different order"
• CONFIRMED the specification was at fault more often than either author
  public/strata/what-two-minds-disagree-about/capstone.json : 6 against 3, 1, and 1 among real disagreements (also true of the fourteen)
• CONFIRMED ten of the eleven shipped as live layers, one held back
  src/content/strata/ : ten slugs have draft: false; the-number-that-will-not-settle has no stratum file
• UNVERIFIABLE the individual fault attributions (who was wrong in each adjudication)
  the engine outputs in research/_wave-2026-08-24/out/ are the final, repaired ones : the pre-repair outputs that would settle the four repaired cases are not kept; the page itself says these are judgements

What was done

• [fixed] headline "84 resolve into 14 distinct disagreements ... four nobody's" rewritten in the dek, meta description, og/twitter descriptions, JSON-LD, figcaption and static figures paragraph: fourteen adjudications, eleven real disagreements (seven in the five still-differing layers, four repaired earlier), three records of full agreement; real split 6/3/1/1. Dated correction paragraph on the page; README and ground note corrected
• [fixed] verifier checked only the count of fourteen; it now derives the eleven, the 7/4 split and the 6/3/1/1 split from the snapshot and requires the page to say so, and requires the old phrasing gone outside the correction note. Dates (YYYY-MM-DD) are now excluded from the orphan-number sweep so a dated correction line does not read as a stray figure
• [declined] (reader) the verifier reads data/capstone.json, which /checks refuses to publish. Not a fault in what the page claims: the page's placard already says running the check "needs the repository around it", and the same JSON is served beside the page at /strata/what-two-minds-disagree-about/capstone.json