# T571 build evidence

The coordinated skill manifest is
`agents/runs/20261028T042605Z-build-skills/manifest.json`.
Both archived versions were read, compared byte-for-byte, and have SHA-256
`e37efb7e2ededf6c1d7134c673f152f648f26e0267c556aebdf3ec65765d8b4c`.
The accompanying `skill-evidence.json` records both versions and their hashes.

## Numerical result

270 positive volumes, both real and complex fields, every 2 <= N <= 16 and
1 <= r <= N. Gamma arguments are integer or half-integer and are reduced
exactly. Each result is a nonzero algebraic factor times a positive integer
power of pi; none is an exact rational concealed as a decimal.

The production evaluation uses 397-bit Arb balls and retains at least 389
relative accuracy bits over the complete selection. Values are serialized to
100 significant digits, each denoting an interval of radius one last-place
unit. The longest stored value is 107 characters, including scientific
notation. The longest working-ball representation is 140 characters. The
standard dry-run measure is 52.4 KB using its parameter-key spelling; the
checker measures 50.6 KB using tuple keys. Both are measurements of the same
270 values, with different key overheads, and are far below the soft limit.

All of the following passed before filling the table, and the numerical
checks passed again on the values read back from the website:

* Determinant and Pfaffian controls include nontrivial positive and negative
  answers, a singular determinant, and a Pfaffian requiring a pivot swap.
* The r=1 and r=2 exponential moment integrals have known answers 1, 1, 2
  for (beta,d,r)=(1,0,1),(1,0,2),(2,0,2).
* All 135 complex eigenvalue integrals are computed independently by exact
  factorial-moment determinants, and all 135 real integrals by exact
  skew-moment Pfaffians, with the odd-rank border included.
* Every generator ball is finite, positive and narrow. Every independent
  560-bit ball retains at least 480 bits. Both balls lie inside the actual
  100-digit serialized interval, not merely overlapping it.
* The four N=2 values agree with elementary disks, circles, balls and spheres
  of radius 1/sqrt(2).
* All 30 pure-state cases agree with the corresponding projective volumes;
  all 30 boundary/interior ratios agree with D*sqrt(N*(N-1)); all 15 complex
  full-rank cases agree with the separate closed expression (4.5).
* The table's Programs snippet agrees with all 270 independent values.
* Generator.verify(sample=None) reports 270/270 matched, 0 differing,
  0 missing, 0 extra.

The independent integral calculation is different from the production gamma
product, but shares the stated metric and geometric decomposition. It is a
builder check, not the separate independent critique required by the campaign.
No Monte Carlo estimate or fixed-precision value is used as a proven bound.

## Audit and state

The site audit reports one structural advisory: `field` has two named values.
It reports no broken link, citation, snippet, tag, precision, or prose finding.
This advisory was assessed rather than hidden: real/complex specifies the
scalar field of the density-matrix space, and each row holds the same
intrinsic Hilbert–Schmidt volume. The declaration and number header name that
one quantity. This is the skill's parameter-as-argument case, as required by
family #227, not two named quantities or a choice between HS and Bures metrics.
The advisory remains mechanically present; `clean` is therefore false.
The separate critic should assess this reasoning independently.

T571 remains unpublished and unoffered. The controller owns independent
critique, repair, and any later offer; no review or publication action occurred.

## Reproduction and files

Download `generate.py`, `checks.py`, `check.py`, and `argument-selection.md`
from the draft. Run `sage -python checks.py` for the private checks, and
`sage -python checks.py --stored` for the readback checks. An accessible
private draft requires a key in NUMBERDB_API_KEY. Run
`sage -python generate.py` for the generator's all-entry verification.
The campaign used `agents/sage.sh` for every Sage computation.

Primary sources and download hashes are preserved in `sources.json` in the
run artifacts. Source locators are in the table and argument-selection.md.
The full source PDFs, scans read, logs, website document, and API audit output
are preserved beside this report. No pipeline, skill, or other source file was
edited, and no Git operation or deployment was performed.