back to table · edit · history · where entries came from · files · download
3869 bytes, as of the version from 2026-10-30 06:27 (current). Recorded here, not run.
# T619 repair handoff, run 20261030T061717Z
Both findings in agents/critiques/T619.md were still present in the live API
document and the rendered page. Both are repaired in the same private draft.
No values, parameter identities, parameter declarations, completeness claims,
argument-selection justification, or original generator sources were changed.
The Programs example is now a self-contained Sage evaluator of the analytic
n=0 and n=1 cases for integer ell>=2. It returns exact rationals at n=0 and
uses Arb cylindrical Bessel functions at an enclosed pi for n=1, increasing
working precision until the requested relative accuracy is obtained. Examples
compute the additional degree ell=4. The related-table description now ends
at "analytic index-one solution".
The formulas were checked against Yip and Leung section 5.3, equations (28)
and (33); repair-sources.md gives the derivation and source-access details.
Executed the complete replacement program using agents/sage.sh. Independent
hypergeometric evaluations at 180 and 220 decimal digits check n=1 at ell=2
through 30, plus 50 and 100. Exact exterior matching checks all 31 n=0 cases.
All four stored analytic values are consistent with these calculations,
including containment of the n=1 computed balls inside the stored +/-1-last-
place intervals. The original n=0, ell=4 refusal was reproduced, and the new
program returns exactly 1/4. Invalid-domain inputs are rejected. The minimum
measured accuracy was 412 bits. The 16 unchanged nonlinear entries were not
recomputed by this repair; no new claim of their full-precision verification
is made. All 20 stored identities still match the generator's enumeration.
API write succeeded with reviewed=false. The readback exactly matched the
proposed document. The post-edit audit returned clean=true and no findings.
Chromium rendered all 71 math containers without math errors, JavaScript errors,
or page-width overflow. The repaired program and shortened relation appear.
A fresh independent critique is REQUIRED: Programs is inside the campaign's
selection fingerprint. The original independent sidecar has not been written,
updated, accepted, or replaced by this repair. The controller should recover
this same T619 draft. No offer, publication, human-review action, replacement
draft, or completion receipt was made.
Old selection SHA-256:
69d483e52b58438d73c4ed2390e28ce03b1cadf6db7db625ea7a0d3822c889a4
New selection SHA-256:
63a638a481605a4b4a09eaec17ed322c75557757bfbe06b3f76e5681da0b407a
Unchanged generate.py SHA-256:
5afb852869c82e56a3b1e8261727b9861ced0b460134577869ce9d7f579b13b2
Unchanged tidal_solver.py SHA-256:
9c218d1f51a2851f674afec3eeae3f02b2a523b620df8bafca74c28da8ea8491
Controlling manifest:
agents/runs/20261030T061717Z-repair-skills/manifest.json
Manifest SHA-256:
f3ffe644f05a2d8ba38978976adde3737ab5de3d1ebeb214066f311233b3080e
Archived local skill SHA-256:
28ecbeeccb0c5e5d6eb541b5a5b0f88aec47b154374d909085116d34f5629199
Archived live skill SHA-256:
28ecbeeccb0c5e5d6eb541b5a5b0f88aec47b154374d909085116d34f5629199
Both copies match the source snapshot canonical skill. The manifest and hash
evidence are also attached as repair-skill-manifest.json and policy-evidence.json.
Reproducible evidence: program-repaired.py, repair_checks.py,
document-before.json, policy-evidence.json, repair-numerical-evidence.json,
and the retained generate.py and tidal_solver.py attachments. Download them
into one directory and run `sage -python repair_checks.py`. The check does not
write to the website. Numerical checks ran with Sage 10.9, Python 3.12.3,
and mpmath 1.4.1. Browser versions were Chromium 156.0.8078.4 and Playwright
1.64.1. The local evidence directory also preserves failed exploratory probes,
all scripts, API responses, before/after documents, screenshots and fingerprints.
All started computations finished.