# Rigid conifold local polynomials: checked research Selected proposal #348 contributes only to wanted #261. The supplied initial message cites both exact issues; the revision history preserves those links. The live #261 discussion distinguishes this queued conifold work from the existing smooth, analytic and offered K3 families; broader Hodge requests are coordination context, not additional origins. No request is fulfilled or closed by this build. Checked primary sources: Long–Tu–Yui–Zudilin, arXiv:1705.01663v4, Theorems 1–2, Table 1, Tables 3–4 and Section 5.2. They fix the canonical rank-two weight-three constituent at z=1, its modular correspondence and the trace correction H_p=a_p+chi_A(p)*p. The source-model/template crosswalk is in the stored document and make-document.py. BCM arXiv:1505.02900v2, Theorem 1.3 and Assumption 1.4 give the rational-data character-sum route, including nonsplit primes. Source HTML/hashes are in source-manifest.json; concise inspected passages are in ltyz-passages.txt. These are checked source statements, not the builder's new universal identifications. Normalization: det(1-T Frob) with geometric/cohomological Frobenius, constant term one, determinant p^3, no additional twist. The lower-weight character belongs only to verification metadata. The natural unramified prime domain is distinct from the practical p>5, p<=31 selection. All fourteen classified templates and all eight selected primes are included. argument-selection.md justifies each free parameter. No convention alternative remains unresolved. Refreshed lightweight catalogue: 760 accessible records, one paginated pass, with metadata of plausible candidates inspected before creating T764. T674 excludes z=1; T679 and T680 store different analytic quantities; T683 covers five associated forms within its existing level range. Published companions were preserved. Both correctly formed numeric polynomial lookup controls returned zero published matches. Earlier unsupported polynomial= requests were discarded; see numeric-discovery-corrected.json. already_here and already_asked were rerun, and all four proposed tags resolve exactly. Checked computation: exact truncated sums modulo p^3 and signed recovery from the Weil bound; 112 native integer-coefficient quadratics. Exact PARI newforms selected by independent LMFDB coefficient prefixes agree through index 99; all 112 traces agree with separate 384-bit BCM sums. Unused primes 37 and 41 give 28 further checks; eight all-half convolutions and sign/character controls pass. All four zero traces are retained. check-results-offline.json and check-results-stored.json contain actual results and versions. Numerical evidence is finite computation, not a new proof of modularity. LMFDB trace downloads/API and one exact displayed q-expansion transcription are preserved in newform-reference.json with URLs and hashes. Challenge HTML was never parsed as data. The level-216 transcription is independently reproduced by PARI through index 99. Sage 10.9, PARI 2.17.1, Python 3.12.3, numberdb 0.1.10. The displayed Programs example is tested in a fresh process; an earlier QQ-based example exposed a missing startup import and was replaced by the already checked modular route. All failed-run evidence is retained. Skill manifest: ../20261119T014035Z-build-skills/manifest.json. All resources were hash-verified; the local/live complete bundles match SHA-256 b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2. Pending: fresh independent scientific/argument-selection critique, bounded repair if warranted, and family review. Full authenticated visual inspection also remains pending; rendering-findings.md states the request-limit and inconclusive screenshot limitations. The sole audit finding is definition length. No independent verdict, offer, publication, review acceptance, queue completion or controller receipt was written by the builder.