build-handoff.md

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

3995 bytes, as of the version from 2026-10-31 12:22 (current). Recorded here, not run.

# T638 build handoff

Built **Chandrasekhar H-function moments for conservative isotropic scattering**
as https://numberdb.org/T638, a private, populated draft. No offer, review or
publication action was taken. The controller must run the separate independent
critique and repair stages before offering it.

The draft contains all 63 raw moments for integer dimensions 2 through 10 and
moment orders 2 through 8, in order (d,n), at 100 significant digits. Its
normalization is alpha_0=2 and alpha_1=2/sqrt(d), with the physical positive
conservative H-function, isotropic angular weight and c=1 exactly. The generator,
independent checks, data, argument selection and numerical evidence are attached
to the draft. The raw downloaded sources and all logs remain in this workdir.

The family was read with `python3 agents/queue.py show 255` before mathematical
work. Fresh `already_here`, `already_asked`, value-lookup and tag checks found
no duplicate family. T637 and the ingredient tables T348, T9 and T7 were read;
their actual slugs were resolved, and the draft links their specific mathematical
relationships. T637 was not modified. The corrected lookup results are in
`screening-fix.json` and `resolved-links.json`; the first low-level lookup used
the wrong JSON shape, and that failed request was not treated as an empty result.

The finite extent was independently reassessed during the build, then retained:
both free parameters are genuine discrete indices. `argument-selection.md`
explains breadth, bounds, omissions and conventions. There is no exploratory
continuous grid, extra quantity or campaign-excluded statistical family. The
independent critic must assess this justification and issue its own required
arguments sidecar. This build has written no critique acceptance.

The governing skill manifest is
`agents/runs/20261031T115431Z-build-skills/manifest.json`. Both archived copies
match SHA-256 `a2b28d9953264d7b0ca62663785c2952190c5be6d6d3597fd695d9a54d606278`.
It is copied to the workdir and attached as `skill-manifest.json`. The artifact
policy, including no offer by the builder, governed this run.

Validation completed:

- Dry run: 63 entries, longest 103 characters, block 8.8 KB; all checks pass.
- 150/200 working-digit computations: maximum absolute change 1.26e-151.
- Independent Fock/direct angular quadrature: all 63 agree within 7.55e-15;
  lower-moment and nonzero quadrature controls pass.
- Published 3D moments, exact low-dimensional formulas, moment recurrence and
  the relation to T637 pass.
- Database `verify(sample=None)`: 63/63 matched, none differing/missing/extra.
- Identities and independent quadrature were rerun against database values.
- Final API audit: `clean: true`, no findings.

Rigour is honestly `heuristic (agreement-checked)`. The high-precision agreement
supports the 100-digit precision claim; independent double-precision quadrature
checks the mathematics but does not certify all 100 digits. No rigorous
quadrature bounds were established. See `numerical-evidence.md` and JSON.

The sources and family settled the mathematical convention. Build choices were
endpoint/cancellation treatment, working precision, numerical methods and the
practical finite cutoffs. An audit led to a shorter definition and explicit
installation instructions. The final tags are `physics`, `integral` and
`special functions`; the operational reason for dropping family-proposed
`optics` and `moments` is recorded in `environment-notes.md`. No skill or
pipeline source was edited, and no new tag is proposed.

`build-fingerprint.json` binds the final selection and `generate.py`; the sole
selection-defining generator source has no imported local dependencies. Its
SHA-256 is `8f708c81b3e88bc1df97720454cdceb34bdf50c1988fa1d515f673afbcda5ed0`.
The critic should obtain its own current fingerprint, check all attachments,
and write the required independent sidecar. No queue issue was closed by this
build, and no batch-exhausted marker was created.