repair-handoff.md

back to table · edit · history · where entries came from · files · download

3650 bytes, as of the version from 2026-10-28 19:40 (current). Recorded here, not run.

# T588 repair handoff

The existing private draft https://numberdb.org/T588 was repaired as zeta3
through the authenticated API. No table was created, offered, published or
reviewed. The controller should retain and recover this same draft.

Four Similar tables links now use their target tables' actual titles. The first
ker occurrence in Comments is linked to its function-family table. The original
Definition is restored: adding its proposed family links introduced a live
audit finding, so that part needs human resolution of the audit/skill conflict.
The separate `digits: 100 (Unknown key)` renderer defect remains documented in
site-renderer-observations.md. The per-finding report is
`agents/critiques/T588-repaired.md`.

All twenty numbers, their order, parameters, function definition, formulas,
Programs, completeness note and precision/rigour claims remain unchanged.
The four reviewed attachments remain byte-identical: generate.py, checks.py,
argument-selection.md and root-counting.md. The generator and reproducible
checks were already attached and were not rewritten. Additional repair evidence
is attached separately so their history and hashes are preserved.

I reran the unchanged attached checks.py against the fresh API snapshot using
`agents/sage.sh repair-checks.py checks.py T588-before.json`. All twenty written
100-significant-digit intervals passed the explicit-tail real-series endpoint
sign checks, displaced-interval negative controls, ODE residual and phase-sign
checks. The known closed form and origin controls passed. This is a rerun of
existing checks, not a new independent critique. Sage 10.9, numberdb 0.1.10,
mpmath 1.4.1. Reproducible wrapper: repair-checks.py; full results:
repair-numerical-evidence.json.

Live target documents and all five target pages were checked. Before and final
HTML from the live preview renderer was inspected in Chromium 156.0.8078.4 with
Playwright 1.64.1. The changed sections have zero MathJax or page-script errors.
Direct /T588 remained HTTP 404, so the full signed-in page is not attested.
The final authenticated API readback equals the final proposed document;
`T588-audit-final.json` reports `clean: true`, `findings: []`.

A fresh independent critique is required because the Comments link changes the
campaign's prose-sensitive selection fingerprint, although its mathematical
meaning and arguments are identical. The original `T588-arguments.json` has
not been edited or replaced. Its SHA-256 is
`1692bf544997463643550bf00c39564b70c6cb7eff901189ec8ce0b8476d7464`.
Old selection digest: `2b3c734058d9e1b96decab145ce6d23f5e7493a25b03b657aaea3ad8598543cc`.
Final selection digest: `3cf53d2c9baa2ed86b4b3692670f5ad371960718331d66703796a751241093ac`.
The generator-file hashes still match the independent sidecar exactly; see
fingerprint-final.json. The repair does not attest a new acceptance.

Instruction authority: `agents/runs/20261028T193331Z-repair-skills/manifest.json`.
Both archived versions were read; their manifest digests were recomputed.
Local SHA-256: `e37efb7e2ededf6c1d7134c673f152f648f26e0267c556aebdf3ec65765d8b4c`.
Live SHA-256: `e37efb7e2ededf6c1d7134c673f152f648f26e0267c556aebdf3ec65765d8b4c`.
The canonical source file and a later direct /skill fetch also match.
See skill-evidence.json and the same manifest/hashes in numerical evidence.

All working files and evidence are in
`agents/runs/20261028T193331Z-repair-work`. The numerical results, failed audit,
both revisions, final diff, preview HTML and screenshots are retained. No
source, prompt, skill, independent review sidecar or controller state was edited.
No computation remains running.