back to table · edit · history · where entries came from · files · download
3075 bytes, as of the version from 2026-11-02 11:01 (current). Recorded here, not run.
# T650 repair handoff
T650 remains the same private draft. Only its four Formulas items changed:
line breaks separate the integrals and periods; the hypergeometric expression
uses two aligned terms with the common coefficient distributed exactly;
the elliptic curve equation is multiline and its j-invariant is separate.
All ten stored values, identities, radical labels, parameters, family definition,
completeness note, program, and previously reviewed attachments are unchanged.
No draft was created, offered, published, or reviewed.
The instruction authority is
`agents/runs/20261101T122610Z-repair-skills/manifest.json`, SHA-256
`2ed18a5ba549629b2149341ece7834c3c10445d5345622bccc42c66178a837ba`.
Both archived skill versions, the canonical snapshot and a current live fetch
match SHA-256 `a2b28d9953264d7b0ca62663785c2952190c5be6d6d3597fd695d9a54d606278`.
The local/live hashes are recorded separately in `skill-provenance.json` and
`repair-numerical-evidence.json`.
The before and after screenshots and DOMs are in this run's work directory.
Live Chromium 156.0.8078.4 measurements at viewport widths 1440 and 390 are,
respectively, 648/648 and 359/359 pixels (available/content width); before the
repair they were 648/726 and 359/724. No page or MathJax errors were observed.
The entire changed section was read and its mobile screenshot inspected.
`repair_checks.py` proves the redistributed hypergeometric coefficients equal
in an exact rational-function field and checks that both parameter matrices
are identical; it verifies the curve's j-invariant symbolically. Numerical
controls at t=24 and t=18-6sqrt(3), computed at 135 and 185 decimal digits with
radicals and cube roots rebuilt at each precision, agree to 135.60 and 136.32
digits and differ from the stored midpoints by 0.34957 and 0.46898 ulp. These
are agreement checks, not certified enclosures or a new independent review
of all ten values. The original heuristic rigour remains appropriate.
Run `sage -python repair_checks.py` with `document-before.json`,
`document-candidate.json` and `skill-provenance.json` beside the script.
The campaign ran it through `agents/sage.sh`; the full output is preserved.
The post-write audit is identical to the pre-write audit: nine param-latex
warnings. The labels are exact radical input identities permitted by the skill,
so they are retained. The digits:100 unknown-label problem is still present
and belongs to the site's renderer; see `environment-findings.md`.
A fresh independent critique is required. Formulas are part of the selection
fingerprint even when mathematically equivalent. The old sidecar
`agents/critiques/T650-arguments.json` was not edited or accepted by this repair.
The read-only gate correctly reports stale selection_sha256. The controller
must recover T650, not create another draft. The new fingerprint is
`757921372c0fae4a6bf085a63fc1bd1fa602990afbf1adb5b1a6694d13cff8a3`;
all four reviewed attachment hashes remain identical. The document repair
revision is `598b96f78daff28b791005ac47e98d16bc891fd0b27682189a6df0cc59bacb04`.