# Build handoff: T765

Private populated draft: https://numberdb.org/T765
Title: Primitive even modular periods of rational newforms.
Proposal: numberdb-data#349; originating partial wanted request: #262.
The draft remains in progress, has not been offered, and has not been reviewed
or published. The proposal claim remains for the controller to settle through
queue.py after the required independent validation. No other table was edited.

## Quantity and coverage

The upward identity-coset Eichler integral has even part
r_f^+(X)=(r_f(X)+r_f(-X))/2=i*Omega_f^+*P_f^+(X). The primitive integer P has
coefficient gcd one and positive leading coefficient, with no rescaling of X
or dropped factors. The Fourier normalization is a_1=1, trivial character,
rational Hecke field, even k>=4 and exact level N. Signs are preserved.
Parameters are ordered N,k,i,normalization. The index counts all real embedded
newforms before restricting to rational fields. normalization is upright in
its display and has the invertible conventions real-scale and path-scale.
The latter stores i*Omega with an exact zero real component.

There are 18 forms and 36 entries, each at 100 significant digits: all rational
forms at level one in even weights through 26, and all rational weight-12
newforms at exact levels 2 through 10. The noncanonical finite bounds and
complementary slices are justified in argument-selection.md. No extra table
or odd period was built; no existing published shape was modified.

## Completed build checks

- Verified matching archived local/live skill resources. Coordinated manifest:
  agents/runs/20261119T212714Z-build-skills/manifest.json, also attached as
  manifest.json, SHA-256
  7629363ce07d2e1340528294fe639e6fc7a5b18da808ec52558687fc1db91534.
  Bundle SHA-256:
  b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2.
- Refreshed authenticated catalogue, title/text searches, already_here,
  already_asked, sample value lookups and exact tag registry lookups. No
  duplicate family was found. Source and defined-period lookup controls are
  retained. The four applicable suggested tags are unchanged.
- Reconstructed every exact primitive even polynomial and matched live T695;
  exact dimensions agree in all 21 spaces. All 33 embedded-form indices agree
  between independently constructed Sage/PARI forms and the companion roster.
- Main PARI period evaluation at 160/220 decimal working digits plus 32 guard
  bits; every raw coefficient is consistent with the exact shape. Rigour is
  honestly heuristic (agreement-checked), not a generic certification of PARI.
- Independent split Mellin sums use exact Sage Fourier coefficients, 450/600
  terms at 150/190 decimal digits, ball arithmetic and a positive Fourier-tail
  majorant. The exact Fricke sign accounts for Sage's N^(k/2-1) operator
  normalization and agrees with PARI. Every even coefficient is checked,
  including exact zeros. Every nonzero coefficient-derived scale lies inside
  its stored 100-digit interval; worst residual is below 0.47 stored ULP.
  The narrowest independent scale has 520 relative-accuracy bits.
- Known nonzero integral controls, an integer Delta eta product, its primitive
  shape, wrong signs, doubled scales, NaN and wide-ball controls passed.
  Six completed level-one Fourier/Mellin checkpoints were reused after a
  higher-level-only operator normalization correction; dependency hashes,
  old checker source and the exact reuse boundary are preserved.
- Existing dry_run.py/check.py passed. Measured stored block is 7.27 KB;
  native-value dry run is 20.3 KB, largest 508 characters. No suspicious exact
  decimals, Python floats or nonfinite enclosures passed the checks.
- Stored readback equals every intended entry; verify(sample=None) reports
  36/36 matched, none differing/missing/extra. Fresh independent Mellin checks
  passed against all stored decimals. Parameter order and exact zero phase
  components were checked.
- Initial history message was read back exactly, preserving #349 and #262;
  codex-cli / gpt-6.1-sol attribution is visible. Run configuration is xhigh.

## Audit and presentation

The authenticated audit has two advisory findings, assessed in
 audit-assessment.md: two normalization names (explicitly one invertible phase
convention), and the definition's length (two necessary defining sentences).
Neither was concealed by retagging, splitting or cosmetic editing.

All six intended Similar tables relations appear in the server-rendered
preview. All 36 entries and all original document sections were inspected in
a complete local composition of exact server-rendered previews. Screenshots
are preserved; the number scroll box was expanded locally to expose all rows.
The unchanged public font was loaded explicitly to resolve a browser font
failure. No math-render or browser page errors remained.

Direct authenticated private-page inspection is still pending: this run has
an API key but no browser session, so the direct webpage returns 404. A
single full-document preview exceeded the server URL limit. The preview
composition is explicitly documented and is not asserted to be a visit to
the direct private page. The critic/controller should finish that check with
an authorized browser session. presentation-evidence.json records the scope.

## Evidence and next stage

Working evidence directory:
/home/ubuntu/numberdb-campaign-20261116-sol/repo/agents/runs/20261119T212714Z-build-work

Core files attached to the draft include generate.py, forms.py, checks.py,
argument-selection.md, research.md, manifest.json, roster.json, fourier.json,
numerical-evidence.json, exact companion snapshots and deterministic checks.
Postfill, audit, history and presentation evidence accompany this handoff.
Final attachment hashes and live metadata are retained in final-metadata.json
and final-uploaded-files.json in this evidence directory.

A live argument fingerprint read was rate-limited while finishing the build;
its failed output/log are retained. The critic must use the actual fingerprint
command against the current live selection and both generator source files:

    python3 agents/argument_review.py fingerprint T765 --generator generate.py --generator forms.py

No scientific verdict or independent argument-review sidecar has been written
by the builder. A fresh independent critic must check the mathematics,
selection, final live revision and attachments, and resolve direct-page
inspection. Only the controller may advance the draft through its guarded
workflow after independent validation and any warranted repair.
