back to table · edit · history · where entries came from · files · download
5668 bytes, as of the version from 2026-11-11 07:57 (current). Recorded here, not run.
# T713 build evidence and handoff
Built Generalized log-sinh integrals as private draft T713, https://numberdb.org/T713 (slug Generalized_log-sinh_integrals). This build does not offer, publish or review it. A fresh independent critic must decide scientific validity and argument selection; this report is not that verdict.
98 real values, 100 significant digits, rigour proven by the finite-series enclosure and integrated remainder in computation.md. The exact finite selection and all free-parameter rationales are in argument-selection.md. Parameter order is n,k,x. All 98 native returns were RealBall; minimum relative accuracy 547 bits at 614-bit working precision. No float input or uncertain exactness occurred. The repository dry_run.py/check.py passed, with longest ball text 204 characters and total inspected block 23.9 KB. Serialized entries are smaller.
Independent mpmath quadrature checked all 98 entries at 140 decimal digits. Four difficult entries were repeated at 175 digits. Worst relative midpoint discrepancy was 1.08065433924e-140. Independent dilogarithm controls covered all three endpoints, direct alternating inverse-binomial sums covered all eight first moments at 2logphi, and the zero-logarithm and zeta known-answer controls passed. Wrong signs, displaced intervals, excessive widths and NaN were rejected. checks.py reads exact rational endpoints; stored-check-results.json checks the actual 100-digit stored intervals. It reuses the unchanged independently computed quadrature values from check-results.json, with program hash binding, instead of repeating that calculation. Every stored interval contains its independent numerical value. This quadrature is numerical evidence, not the source of the proven label.
Reproduction:
- `agents/sage.sh offline.py generate.py agents/table-build/dry_run.py agents/table-build/check.py` (the attached deterministic log preserves the actual run; offline.py is in the run artifacts).
- `python checks.py offline-rows.json check-results.json` with mpmath 1.4.1.
- `python checks.py stored-rows.json stored-check-results.json --reference check-results.json`.
- `cat "$NUMBERDB_KEY_FILE" | agents/sage.sh verify.py generate.py`: verify(sample=None) matched 98/98, with no missing, extra or differing entries. verify.py and verify-complete.log are retained in the run artifacts.
- Generator default command `sage -python generate.py` verifies; `--publish` fills an authorized target and does not offer or publish a draft. Account key and structured provenance are caller configuration.
Policy evidence: ../20261111T011141Z-build-skills/manifest.json, also attached as skill-manifest.json. Both archived bundles were hash-validated and match at d6661c2f21613524889fa3a57d3c95f8330344ca62e5a856211c08a282b7dbab. Core and only relevant references were read from skill-local. The mathematical passages inspected are BS §5.2 eqs. (33)-(35), Example 11; CGLXZ Definition 2.1, Lemma 2.2 and §4; DK eqs. (1.1)-(1.3); and DLMF 4.36.1 in the web reader. Raw source HTML and concise extracted passages remain in the run directory. Direct DLMF downloading failed, but its displayed product was read through the web tool.
Discovery: one complete lightweight catalogue planning pass, local filtering, already_here/already_asked, synonym searches, three numerical api/lookup searches, current metadata/full documents for plausible companions, and a refreshed nearby draft listing. No existing family matched. The numerical pi^2/10 search match was a different Chern-Simons context. The wanted origin #263 and proposal #317 were checked; their explicit contributes-to mapping is preserved in both creation and fill revision messages. No provenance issue text appears in the mathematical prose. Existing tags were used; the proposed polylogarithms tag was absent and remains a family-level suggestion.
Audit: authenticated /api/table/T713/audit reports one heuristic about the three named x keys. This is a false positive: x has type R and denotes exact real endpoints of the same Lsh function, explicitly reconstructed in the generator and defined in the prose. They are not separate named quantities or convention selectors. No schema/link/tag/precision defect was reported. The audit finding is retained without renaming stable keys or splitting the table.
Presentation: public canonical /T713 returns 404 because the draft is private and this runner has no browser login session. The exact stored document was therefore rendered in sections through the site's public non-saving /preview route. All prose, reference and formula fragments returned 200 with no renderer alerts; six representative rendered pages were inspected with Playwright, including the full n=10 value block. The site's MathJax compiled all 74 TeX fragments with zero errors, and rejected an intentionally malformed fraction. browser-results.json, screenshots, HTML and scripts are preserved. The sources and companion HREFs resolve in the audit. Revisions were read back publicly at /revisions/T713; the exact proposal and wanted citations and codex-cli/gpt-6-astra attribution are present. A later sibling can link the duplication boundary; no sibling was created or edited here.
Source-of-truth and limits: the website owns the draft and its attachments. Full source archives, checkpoints, API responses, exact stored document, screenshots and a SHA-256 inventory are in the build work directory. No independent verdict or controller completion receipt was written. No published table was edited, no tag was added to other tables, and no queue issue was closed. No unresolved numerical discrepancy remains from the build checks.