handoff.md

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

5886 bytes, as of the version from 2026-11-18 22:18 (current). Recorded here, not run.

# Build handoff: T753

Populated private draft: https://numberdb.org/T753
Title: Laurent coefficients of the secondary zeta function at 1.
Author account zeta3; run 20261118T135803Z, gpt-6.1-sol, xhigh campaign setting.
The draft remains in progress, unoffered, unpublished and unreviewed. Only the
controller may advance it after fresh independent scientific and argument
selection validation. This report is a builder handoff, not either verdict.

The 42 entries cover each order n=0,...,20 in both conventions c_n and
C_n=n!*c_n. The zero divisor is alpha=(rho-1/2)/i over every upper-half zero
with multiplicity, principal powers and meromorphic continuation; no RH or
simplicity assumption. Parameter order form,n and a single display group keep
each n's coefficient before its derivative without changing entry identities.
Both forms coincide at n=0 and 1. No continuous coupling or excluded quantile
parameter exists. argument-selection.md justifies the full natural family and
the practical finite cutoff.

Generating certification: four source files, 493-bit complex Arb evaluations,
512-node Cauchy extraction at radius 1/2 and a verified interval-arc bound
|F|<4096 at radius 7/8. The explicit geometric alias bound is added to the
output balls. Worst relative radius is below 6.9e-112 before 100-significant-
digit serialization. No decimal or floating-point pattern is treated as exact.
All native samples, original source hashes, versions and dependency code are
retained. See numerical-evidence.md and cauchy-evidence.json.

Completed checks:

- Existing dry_run.py/check.py: types, finite balls, decimal exactness traps,
  imports, prose and sizes pass, 42 entries and 7.9 KB native block.
- An independently derived central-xi resolvent Mellin calculation checks all
  42 entries in two full passes at 170/210 decimal digits with distinct
  quadrature and series depths. This is agreement evidence, not a second
  rigorous certificate. It uses no secondary-zeta evaluator or sibling values;
  shared xi/Hadamard theory and Arb small-t arithmetic are disclosed.
- All 75 stored-value comparisons pass: 42 independent integral results,
  21 factorial relations and 12 published Kawalec Table 9 controls. Exact
  rational endpoints enforce the stored decimal intervals. Known-answer,
  deliberately wrong factor/sign/factorial, off-line mock-divisor, NaN,
  vacuous-precision and float controls pass.
- Fresh generator verify(sample=None) returned 42/42 matched, 0 differing,
  missing or extra. Its subsequent metadata logger failed on a slotted report;
  the actual report and exception remain in verify-stored.log, with a precise
  parsed record and a corrected adapter. No successful check output was
  invented and no value was changed to match a checker.
- Authenticated audit: one assessed advisory about the two form names. The
  canonical invertible-convention exception applies and is already explicit
  in the Definition. See audit-assessment.md; there are no other findings.
- All ten live preview cases pass without math-rendering errors, including
  all 42 stored rows. The four intended Similar tables records appear:
  T752, T733, T15, T36. Screenshots were visually inspected. The direct private
  page needs a login session unavailable here, so exact authenticated API
  read-back sections were rendered through the site's public preview instead.
- The revision history retains the exact initial message citing proposal
  https://github.com/numberdb/numberdb-data/issues/342 and contributing to
  https://github.com/numberdb/numberdb-data/issues/275, with the model recorded.
  No origin is asserted for issue #329 and no wanted issue was closed.

Tags: reused Taylor series, L-function and spectral zeta functions from the
live registry; added the justified absent Laurent series concept after exact
case-variant checks. Taylor and Laurent series name distinct applicable
concepts. No suggested sound tags were removed and no unrelated issue's tags
were imported.

Retained failed experiments are diagnostic, not unresolved mathematics: the
radius-1 bound crossed an evaluator's removable internal singularity, so the
bound circle is 7/8; a Decimal-context bug in a developing checker caused false
normalization disagreements and was fixed with exact digit/exponent parsing;
minor modular-Sage/serializer/report interface failures are recorded. The
selected values and their certified computation did not change to match those
checks. Browser/deployment notes are environment-findings.md; contributor
lessons are an unpromoted artifact proposal, never a skill edit.

The mandatory next actor is a fresh independent critic. It must inspect sources,
independently check the stored numerics and all parameter choices, and produce
its own revision-bound scientific verdict and argument sidecar. The selection
fingerprint at this build is
c342048864e0eaf541719e40c6443bce4e5ac1e22c2a5cdd94d1d16de6f0ad5d;
selection-fingerprint.json names all four live generator dependency hashes.
No builder acceptance sidecar has been written. Final current revision and
attachment hashes are in final-live-metadata.json, after the final source
attachments; the controller should refresh them for its immutable packet.

Skill manifest: ../20261118T135803Z-build-skills/manifest.json, also copied as
skill-manifest.json. Both archived bundles and every resource hash validate;
common bundle SHA-256 is
d6e85d49873fac5bc748d068b0288cf242b0f9aebf3a9a0fff70295019227420.
Only the archived local core and its task-relevant references were read.
evidence-manifest.json indexes durable evidence with actual hashes.

Artifact handoff file for the controller:
../20261118T135803Z-build.table.json, containing only {"tid":"T753"}.
No source snapshot, shared toolchain, skill, controller state or completion
receipt was changed; no commit, push, deployment, review or offer was made.