[petalas]

Summary

PR3462 leaves 93 omitted P104 banks in their initial physical charts after its normalized endpoint correction. This isolated package supplies the exact missing adapters and proves their sharp bank-local native-only tariff.

for each omitted bank b:
    J_initial(b) q
      -> M_left C_rank M_right
      -> J_final(b) q

52 rank-eight calls + 41 rank-nine calls
rank mass = 785
sharp tariff = 52*8^s + 41*9^s, for every 0 <= s <= 1

The proof permits any finite separate nine-axis bank-local Clifford word, arbitrary intermediate frames and free affine quadratic-phase monomials. Pauli B-block rank is subadditive; the direct recipes attain its floor and the concave-power tariff. Cross-bank coupled operations and chart changes performed inside the scalar completion are outside this claim.

The package reconstructs PR3446's 4,027-event/265-input projection from the unchanged original PR2586 inputs. None of its retained events targets an omitted bank. All 53 distinct explicit recipes and their 93 bank bindings are committed. The 41 zero normalized residual rows still require physical adapters.

Credit Hiroki Kumazaki/PR3446, PR3462, Clear1nsight/PR3382/3423, ywz65/PR2586/2497, Arham Tahir/PR2665, the retained Jeremy Kothe Clifford-compiler lineage, PR3416's separate checker, and the original native/geometry/compiler contributors named in README.md. PR3466 and concurrent PR3467 address different entrance obligations. Frozen source bytes and notices remain unchanged. Base and compatibility checkpoint are canonical main 6f100d4bb5083f8c35d96fd95a539811c62f245c, priority revision 6. Canonical claim.

Evidence

• Before: the normalized completion supplies no final chart maps or paid native profile on these 93 banks.
  After: each recipe equals the requested chart map on all 18 signed Pauli generators and an exact 512-entry anchor, fixing global phase. A separate bounded elimination computes every rank floor. Dense Gaussian payload and inverse-restoration checks pass.
• Registered normal and optimized Python 3.12 replay and five behavioral groups pass. Controls reject a global-phase fault with unchanged tableau, wrong rank/endpoint/wrapper phase, damaged receipt/checker, and changed/missing source bytes. The independent small rank oracle covers all 512 three-by-three binary matrices.
• Scoped Ruff, whitespace, exact publication-byte and source-history checks pass. The committed source-only export passes normal/optimized verification and controls with Git unavailable; all 15 exported regular files retain their bytes, modes and inodes. Hosted acceptance is unclaimed.

python3.12 -B scripts/research_package.py --package research/p104-omitted-frame-adapters
python3.12 -B scripts/check_research_commit.py --package research/p104-omitted-frame-adapters --check-source-history

Set TMPDIR to an owned external directory for disposable CLI fixtures. PROOF.md gives the model, algebraic proof, cost ledger and stop decision. SOURCES.json and inputs/ORIGINS.json preserve source pins and the byte-identical separate checker. Runtime needs no pending Git object or network access.

Merge Danger

Door: two-way

Blast Radius: research

Only one new scoped package is added. Monomial implementation costs, the scalar completion and its common-frame readouts, physical routing/ownership, numerical precision, actual paid caller/consumer, full supplier invoice and outer transfer remain open. The original 473 omitted transports are interleaved with other operations and are not a standalone replaceable adapter program. No runtime speedup, complete supplier, improved kappa or selected-frontier change is claimed. The unchanged-profile stop remains; further construction first needs an actual paid caller and complete cost ledger. Independent human and formal proof-kernel review are unclaimed. Lean, Docker and the full selected cold replay were not run.