Preserve current frames in the QQ16 L2 consumer

[ywz65]

The QQ16 L2 read can preserve the full current frame by replacing the old owner-map shift delta with delta + g delta for each actual current-chart involution g. This package constructs all 104 maps, proves their complete 73-bit phase covariance, and uses them in one native word with two explicit endpoints. PR2492's old-frame construction remains correct.

The main endpoint recovers x1_U in the current frame and consumes it in place as y1_U = x1_U - X2_U. Its complete staged invoice includes the 96 consumer ADDs: 192 rank-one + 16 rank-eight children, native rank mass 320, 416 native wrappers, 208 full-cover maps and 7,392 real scalar primitives. All 200 borrowed source arrays are restored; the result occupies the same 96 owned outputs and uses one owned local scratch.

A fixed alternative epilogue returns x1_U in old frame F_L, exactly PR2492's output/source contract. Under the identical positive-child grammar, this costs native rank mass 1,088 versus 1,168, with 608 versus 992 wrappers, the same 208 cover maps and the same 6,624 scalar primitive allowance. The 320-versus-1,168 comparison changes the frame/consumer contract; only 1,088-versus-1,168 compares identical endpoints. These are local invoices, not full physical or supplier costs, and no shortest scalar circuit is claimed.

research/qq16-current-frame-consumer/ contains the arbitrary-array proof, both complete 1,235-event words, exact symbolic owner replay, child kernel checks, all map matrices/translations, 281,008 independently evaluated character comparisons and twelve semantic rejection controls. The zero/singleton/pair evaluations determine the complete modulo-4 quadratic phase; they are not random evidence. Boundary tests cover private cleanup, delayed owner alignment, Git-free read-only export, both committed words/receipt, and source corruption.

Validation on Windows with the standard library:

python3.12 scripts/research_package.py --package research/qq16-current-frame-consumer
python3.14 -I -B -O research/qq16-current-frame-consumer/verify.py
python3.14 -I -B -O research/qq16-current-frame-consumer/test_verify.py

Native CPython 3.12.7 registered replay and 3.14.5 optimized primary/five boundary test groups pass. A separate AI-assisted review derived all 104 correcting shifts from orthogonal row-weight polarization, imported no candidate code, and independently interpreted both literal words. It matched all 480/576 owner-labelled output terms, restored all 200 sources and confirmed every local invoice. No material issue was found. No independent human review, hosted CI, proof assistant or selected supplier cold replay is claimed.

The eight earlier T records must already have been captured and delivered. Whole-cover physical movement, child placement, precision, retention/output lifetimes, the remaining decoder and the global controller are open. There is no storage saving, complete 224-input decoder or new kappa claim. Owned output clears do not authorize clearing borrowed helpers.

Canonical main was freshly fetched at round start, pivot and before review, unchanged at 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4. PR2492 is pinned at 9cc0d4d60819ddd246c9db9549f8ae4a61b68e45; its exact README and adapted code are retained with Git blobs and SHA-256 hashes. PR2470's context head 1b2eee3bf1ec67e68c0b0fa7f47c7b2a66febe56 and PR2492 were REST-verified unchanged before review. Six base inputs, including the original phase/archive, cover contract and scalar cost source, are authenticated before use.

Overlap coverage: complete cached 683 open PRs at 12:30:47 UTC and 1,197 Discussion records at 12:29:05 UTC, supplemented by the latest 50 updated open PRs via REST at 13:14 UTC. PR2511's changed selected-sum observation/early-retention result was checked and leaves this ordinary native L2 interface separate. GraphQL rate limits prevented a newer complete Discussion refresh; this is bounded coverage, not a novelty guarantee. The original consolidation draw and justified native-interface pivot are in TASK.json. Advance claim.

Credit ywz65's PR2492/PR2468 native reconstruction and read machinery, thakrarsagar's PR2470 L2 identity, starship-s's original cut-decoder work, Ali Romman's QQ16 contract and the retained native/geometry lineage. Code is explicitly adapted and source notices retained. New work by ywz65, with substantial OpenAI Codex assistance in derivation, source inspection, implementation, testing, writing and separate-agent review. Apache-2.0.