[Clear1nsight]

Change and mathematical scope

Combine one corrected actual-centre0 edit specification, its independent source audit, the exact rejection of one declared r128 product placement, and independently checked conditional p2 accounting. No changed native word was emitted or executed, no full physical witness or complete caller is supplied, and no new κ is claimed.

Restoring the first literal H8 centre and retuning its helper precompensation/four affected repairs gives G=I-U(I-e0e0^T)R, RU=2I, hence G²=I, under the inherited normalized #4948 source. The backward specification uses actual conjugate frames and coupling −G^T; it is not a claim that the entire producer is a scalar inverse transpose. All eight H1 repairs remain. Local profile is {1:2209,2:412,3:510,4:127,5:40,8:1}, exactly +H8 versus #4948; scalar edits add 714 ADDs per FULL. All 632 original independent owners and the 408-helper inventory remain in the specification.

The declared placement A=G⊕I16, B=G^T⊕I16 in the scaled all-output product P(a,b)=ab/128 on C[y]/(y^128+1) fails the necessary unit-substitution equality at the first permitted probe e1: output coefficient 0 is 143/4 versus −25/4, residual 42. Stop this placement; no other embedding or producer is excluded. The failed initial PORT0/physical-centre identification remains an explicit negative control.

For an unrealized p2 target grant, P={1:4754,2:824,3:1020,4:254,5:80,8:226} and T={4:56,7:14,8:94,9:876,17:224}. The 816 helpers are counted once; 448 autonomous main jobs remain unproved. A uniform derivative enclosure on [.0030217758,.00339999] gives minimum raw margin 12.804117221917453035517808..809. Exactly 12 extra H1 calls, or one rank-12 child, fit uniformly in this additive model; 13 positive-rank calls or one rank-13 child fail at its right endpoint. Finite work, actual endpoint eligibility, both paid tails, compiler reserve and outer transfer remain unfunded/unbound.

Task and overlap

Canonical base 81e9bd69e4a5b00ba9c713a1620e36d576d5522e; task complex-new-helper-inventory, refined to a single source-specification/embedding/conditional-cost audit. Canonical updated checkpoint records source PASS, declared placement STOP and local publication preparation.

Source #4948 86e474038e05f99a76d41f71d81946ad2327e4a4 is preserved through exact original request paths and frozen bytes in SOURCE-REQUESTS/SOURCES; local input copies use same-path retained ancestor pins. No scientific repin or code-expression rewrite occurs. Dillon's active p2 Gaussian-output transport remains a separate task. A fresh all-author created-and-updated PR scan covered activity through 2026-10-23T09:18:17Z (cursor #5098), with direct reads of the canonical #4930 investigation and #5083 comments and head. No exact competing scope was found in that coverage. The canonical main ref remained81e9bd69e4a5b00ba9c713a1620e36d576d5522e at 09:18:52UTC. This is bounded metadata/body/known-claim coverage; arbitrary diffs and unpublished work are not covered, so it is not an exhaustive novelty claim. The distinct #5083 physical cap-two/comparator review is packaged separately rather than included in A's replay.

Validation

Exact package commit: 5cf2cec18084b4ebb4c8e8ad2f97da045fa7fa01.

Exact reviewed and exported package commit: 5cf2cec18084b4ebb4c8e8ad2f97da045fa7fa01.

Python 3.12 standard library. Independent reverse sensitivities derive all eight actual centre output columns and centre0's 45-helper input factor, separately from the author's forward impulses. Exact rational checks bind the repaired coefficient dictionaries, all eight retained H1 calls, backward conjugate-frame edit sites, 632-owner census and G matrix. The product test uses only the elementary necessary unit substitution; no full isotopy-classification theorem or external paper text is required. Positive/negative controls preserve the failed PORT0 witness, scaled identity gate, negacyclic wrap, independent records, wrong output map and inverse-transpose/adjoint distinction.

python3 -B scripts/research_package.py --package research/actual-centre0-source-audit --check-only
python3 -B scripts/research_package.py --package research/actual-centre0-source-audit
python3 -B scripts/check_research_commit.py --package research/actual-centre0-source-audit --check-source-history
python3 -I -W error -B research/actual-centre0-source-audit/verify.py
python3 -I -O -W error -B research/actual-centre0-source-audit/verify.py

Registered normal/O checks pass in 4.297 seconds combined; Git-free read-only official-export normal/O take 1.811/2.220 seconds, with unchanged bytes/modes and identical stdout. Fresh producer/product/independent-p2 receipts match frozen JSON byte-for-byte; the author p2 stdout matches complete JSON. Original manifests, exact source history, committed bytes, whitespace, official export integrity and original science-byte equality pass. Preflight/export removes inherited GIT_* variables in an explicit-root subprocess environment; no shared Git-tooling fix is included. The sole AST adaptation replaces the source audit's Git byte-reader body with hash-checked frozen reads in private scratch. No native/Lean/compiler/global or historical suite ran; hosted CI has now passed at the exact reviewed head.

Hosted CI run 39931528686 passed at 5cf2cec18084b4ebb4c8e8ad2f97da045fa7fa01 on 2026-10-23 at 09:29:44 UTC. Planning/helper checks completed a 400-test suite with 17 expected skips. The sole selected package replay passed normal and optimized modes, preserving the exact declared-product rejection, unrealized p2 target status, unchanged committed inputs, native_word_emitted=false, complete_caller=false and new_kappa=null. Hosted execution corroborates these scoped executable claims; it is not scientific admission or a selected-frontier update.

Credit

Prepared by Chris R / Clear1nsight with substantial OpenAI Codex assistance. Credit Dugongue (GitHub primary author of #4948) and the #4929 lineage, dillona/Dillon for #4933, Cao Yifan/intqwq and native/controller/compiler predecessors, Cinna Davis centre-free/cache/helper-basis work, Astra's changed-producer design and declared diagnostic, the source derivation worker and independent source/product/economic reviewers, and retained native/helper/precision/outer contributors. The elementary unit gate does not claim invention of the general de Groote/Bürgisser–Ikenmeyer classification. Original Apache-2.0 licenses, notices, corrections and failed evidence are unchanged. A historical local claim draft miscredited #4948 to Dillon; its text is retained as provenance and superseded by this explicit primary-author credit.

Open next questions

Declare one different fully specified embedding/product/caller contract before any further search or construction. The source-specification PASS does not deliver the 448 granted main jobs or repair the residual 42 placement failure. Any extra recursive correction must fit its own full bill; a positive conditional raw moment does not fund finite movement/precision, compiler reserve or outer transfer. See README/TASK for the bounded next task and blockers.