# Build handoff: T611

Draft: https://numberdb.org/T611
Title: Volumes of right-angled hyperbolic antiprisms and Löbell polyhedra.
Family: numberdb-data#245. Run: 20261029T183732Z-build.

The table contains 54 volumes with 100 significant digits: every ideal
right-angled antiprism A_n for 3 <= n <= 30 and every compact right-angled
Löbell polyhedron L_n for 5 <= n <= 30. Curvature is -1; n counts base sides;
volumes are for one polyhedron, equivalently its full reflection-group
covolume. The two parameters identify the object, not different invariants.
The practical bound 30 and both discrete families are justified in the
attached argument-selection.md and in the completeness note.

The archived skill manifest is
`agents/runs/20261029T183732Z-build-skills/manifest.json`.
Both archived versions were read and their hashes checked:
- local: 28ecbeeccb0c5e5d6eb541b5a5b0f88aec47b154374d909085116d34f5629199
- live: 28ecbeeccb0c5e5d6eb541b5a5b0f88aec47b154374d909085116d34f5629199

## Evidence and decisions

The queue issue was read first and the live corpus rechecked with already_here,
already_asked, numeric lookup, and existing tags. T85, T131, T175, T222,
T223, T224 and matching manifold-volume contexts were inspected. There is no
duplicate family. Actual companion slugs were resolved; no unbuilt growth
companion identifier was invented. No published table was modified.

The generator evaluates the published formulas using Arb balls at 413 bits.
The narrowest relative accuracy is approximately 122.26 decimal digits.
The standard dry run passed all 54 entries, named imports and exactness;
there are no disguised exact values. Serialized decimals are 101 characters
long, with a measured entries-block estimate of 9.1 KiB before entry comments.
The example Program and Euler incidence identities were checked over all
54 selected objects (supplement.log).

Independent mpmath integration at 115 and 135 digits checks all 54 values.
For antiprisms it integrates log sine; for Löbell polyhedra it uses the separate
Schläfli arccosh integral in Mednykh–Vesnin Theorem 2, dividing the manifold
volume by eight. Three integration controls run first, including two nonzero
Catalan checks. The octahedron equals 4 Catalan, the dodecahedral value agrees
with published digits, and A_4 equals L_6 in volume. Every stored decimal
interval contains its certified Arb ball and the independent numerical value.
The maximum independent difference is 0.499045 units of the last place.

After filling once, Generator.verify(sample=None) reports 54/54 matched,
zero differing, missing or extra. All independent numerical checks were
repeated on the actual Numbers read from the database. The API represents
the records nested under two family groups; this is 54 entries, not two.

A printed source sign error was resolved, not ignored: equation (2) in
Mednykh–Vesnin prints a plus before Lambda(2 theta - pi/2), whereas Inoue's
Theorem 4.1 has a minus. The former gives L_5=2.22788471968..., contradicting
the same paper's independently derived Theorem 2 integral and the known
4.30620760073... value. The latter agrees for every selected n. See
source-review.md and the numerical evidence. Both papers define the same
polyhedron after the explicit factor of eight; this is not an unsettled
normalization or definition.

The audit has one structural family-name warning; it is intentionally retained
with an explicit mathematical response in audit-disposition.md. The audit is
not marked clean. The current skill and family proposal permit both discrete
object families as arguments to the same volume invariant. Independent
critique should assess this, as well as the arguments and numerics.

The existing tags volume, discrete geometry, groups and period are used.
Propose hyperbolic geometry for this table and T222/T223/T224 (also the
planned Löbell growth companion); no singleton tag was created.

## Reproduction and next stage

Website attachments contain generate.py, checks.py, argument-selection.md,
numerical-evidence.json, source-review.md and skill-evidence.json. This run
also retains all downloaded sources, hashes, control failures and their
resolutions, stored evidence, the audit, and dependency versions.

`fingerprint.json` was obtained with the campaign fingerprint command and
binds the live selection plus generate.py, checks.py and argument-selection.md.
No independent critique -arguments.json sidecar was authored by this build.
The next actor must independently assess the arguments and numerics and write
its own sidecar with its own NUMBERDB_RUN_ID. No offer, review, publication,
queue completion, deployment or commit was performed. The controller may
advance the same draft through critique and repair, then decide whether it
can be offered. The unauthenticated API still cannot read T611.