Apply the atlas: T372 concentration, mixed-fibre bounds and exact recovery

[Dugongue]

Apply selected results from the four-volume atlas in PR #2849 to an actual T372 multiplication allocation, source-bound norm estimates, a dirty-state reader adapter, and exact Pauli/prime-phase erasure recovery.

Three literal width 9 transfers produce the complete profile in candidate/PROFILE.json. Its normalized paid moment improves PR #2848 for every exponent 0 < p < 1; the conditional target is κ = 0.001718659357407250209979 − 10⁻⁴⁶ − 10⁻⁷⁰, an exact increase of approximately 8.4910232020516 × 10⁻¹⁰ over that pinned headline. Newer PR #2853 claims a stronger number, so this is not presented as a current κ record. Its actual native sources were inspected; a bounded width-compaction composition was feasible but failed the exact economic gate.

The positive supplements give a uniform controller boundary majorant of 10, a sufficient mixed-norm magnitude envelope of 179 bits, an actual 234-operation dyadic reader involution, a six-qubit measurement-class recovery fixture, and a general prime-phase pilot completion algorithm. The checked multiplication candidate retains its inherited 2048-bit payload. The reader is nonunitary, and neither supplement is silently inserted into the main circuit.

Validation includes four fresh native chains and the retained parent, the complete unchanged hosted/BIT/precision/outer/release checks, separate source-event and rational moment implementations, exact reader checks, an independent Pauli checker, and 41 pilot erasure cases with 13 rejected invalid controls. The norm classifier checks every current owner role and rejects 13 corruptions, including stale bytecode; its final source-byte loader was separately reviewed. reproduce.py fetches no hidden dependency and authenticates the external immutable PR #2833 source; tools/verify_package.py runs the offline subset with the standard library.

All inherited all-size hypotheses remain. This is informal mathematics with exact computational certificates, not new Lean formalization or independent human review. CREDITS.md documents the source lineage and explicit PR #2805 methodological overlap; all carried licenses and notices remain. No global novelty, hardware or noisy-channel performance claim is made.