back to table · edit · history · where entries came from · files · download
2428 bytes, as of the version from 2026-10-24 20:04 (current). Recorded here, not run.
# T464 repair evidence
The only document change removes the final example sentence from Definition.
The graph integral already fixes the normalization, and the special value
remains in the P_(3,1) entry comment and follows from the zig-zag formula.
No values, parameter names, parameter order, comments or formulas changed.
The original numerical generator and its sources remain attached to T464.
`repair_t464.py` reproduces the proposed document from `T464-before.json`.
`check_t464_repair.py` independently evaluates all nine zig-zag entries
P_(ell,1), ell = 3,...,11, using 512-bit Arb single-zeta values and exact
rational coefficients. Run it with SageMath and numberdb installed:
sage -python check_t464_repair.py T464-before.json
In the campaign the completed invocation was:
agents/sage.sh check_t464_repair.py T464-before.json
Every computed ball is finite, has radius below 10^-40 last-place units,
and lies strictly inside the stored decimal's plus/minus-one-last-place
interval. The largest midpoint discrepancy is below 0.416 last-place units;
P_(3,1) differs by about -0.122805 last-place units. A zeta(2) control and
deliberately wrong comparisons test the checker. This does not independently
certify all 852 graph periods or change their heuristic rigour label.
The source identity is Brown–Schnetz, Theorem 1.1:
https://arxiv.org/pdf/1208.1890v2 . Its coefficient at ell = 3 is exactly 6;
the same paper identifies Z_3 as the wheel with three spokes. The downloaded
PDF and both versions of the table document are preserved in this run.
The live skill was fetched afresh. Its differences from the local skill
concern argument selection; the stricter local convention was followed.
The build prompt and the live API documentation were also read.
The fresh rendered-page request returned HTTP 404. Anonymous API access
does not expose T464, while the authorized API does, confirming draft
visibility. The key was used only on NumberDB API routes. The actual rendered
page was not inspected; this remains the critique's unresolved human follow-up.
The write response reports reviewed=false. Readback matches the proposed
single-sentence deletion, all 852 entries are unchanged, and the post-edit
audit reports clean=true with an empty findings list. No publish, review or
offer-for-review action was taken. This repair did not build a new table,
so no build-result table.json is appropriate.