[RossRichardson]

Summary

Replace the fixed R429 source-code/private-repair boundary by nine source-pivot totals after explicitly lifting all 111 targets to FULL. The resulting local word has 12,932 signed ADDs, 2,984 MOVE queries, no private field, and rank mass 5,097. It preserves every arbitrary source/helper logical integer value and implements the target shear over F2.

Include the complete local invoice and a conditional caller analysis: the specified direct L-bank topology has zero rank deficit; an exact eight-bank seeded fanout still fails primitive exponent a>=1/500 under its complete D<=33 invoice. This is one endpoint-interface package with its caller boundary, not a new supplier or exponent.

Scope and limits

The local check covers 651 arbitrary ordinary inputs. The signed target update has nonzero even residuals; the clean shear is a binary statement. Helpers finish physically as F8*h, despite exact logical restoration. Another 111 paid common-FULL ADDs diagonalize the physical binary data block; they are outside the emitted 12,932 count.

All 111 target lifts are counted. The direct D0 identity assumes each bank is source once and target once in distinct calls plus legal helper hosting; aggregate capacity already requires L>=7 and does not prove ownership. The three-seed theorem separately assumes L>=3, unchanged seed profiles and complete D<=33. Native binding at changed cuts, inherited all-modulus compiler behavior, rounding, owner schedules, source access and tape remain separate obligations. The 874 signed prefix bound is not a precision theorem. No selected-frontier or κ change is claimed.

Review corrected two historical statements: inherited ADDs also occur at intermediate common frames, and the new fanout primitive is not an involution. Original bytes are retained with explicit prose/comment-only adapters; arithmetic and proof controls are unchanged.

Test plan

• Complete registered Python 3.12.14 replay at f0625d20bcc175e0865d1d5cf3f7852fe2287679; publication adds only the final integrity review and validation documentation.
• Regenerate and compare every decompressed word byte; all 651 signed/F2 input columns, literal inverse/roundtrip, 429 logical helper returns, full frame/cost ledger and three semantic mutations.
• Independent dense signed operator and all 5,328 dyadic ADDRESS columns; faithful 16x16 caller representation, all 16 prefixes, nine-variable fresh factors and actual idle windows.
• Exact all-L pressure certificate 21108762/609725>33, with sensitivity controls that make no positive changed-supplier claim.
• All 19 retained inputs authenticated; full-field receipt comparisons; source/receipt/type/path-collision/cache controls, gzip-container equivalence and truncated-word rejection. Controls pass normal/-O through the real launcher and from a committed export.
• Metadata and source-history preflight at b1204d8ae309eba8b2808a3f3e2d1ce9de3a2c71; independent AI mathematical and integrity review. No human or formal-kernel certification.
• Automatic exact-head GitHub CI 39243446023 passed the complete registered replay and unchanged-input gate; no manual rerun requested.

Sources and credit

Scientific pin: 20ecf2891d80b5acf18d980b2164fb44ea7faa90. The predecessor canonical word is exactly PR3645, whose helper deletions, minimality and cap audit are prior work. Sagar Thakrar supplied the R438 transport; Ali Romman, DreamingOfClouds and eumemic supply the underlying BIT/source-code lineage. Related mutable-source tours include Hanson Sun/bbMisk PR1522.

James Reed's PR3850, based on Gonzalo-Mier PR3772, prices a different retained FULL-target schedule with L-1 banks and negative deficit. The proof distinguishes that model from the L-bank direct topology here. Current overlap was refreshed before submission; a search snapshot is not a novelty certificate.

Prepared by Ross Richardson with substantial OpenAI Codex assistance and separate AI reviews. Original notices retained; Apache-2.0. Exact source and provenance maps are included.