Fuse full QQ16 recovery from 336 to 224 native calls

[Clear1nsight]

The full PR #2707 recovery word transports every future-logger field into and out of its query frame, although the logger reads 104 ports and writes only eight. This contribution replaces that block with an owned rectangular correction and preserves all 224 original outputs in F17.

The complete local bill changes from 336 to 224 H1 children, from 672 to 448 native wrappers, and from 464 to 352 cover maps. It adds 1,344 scalar allowance (27,184 total) and 112 owned full-cover temporary arrays, all returned zero in identity frame. All 224 current X/Y fields stay borrowed and unchanged. Owner-private peak remains 16. The literal removals are the 104 query-source return transports and eight response-entry transports; the original predecessor and F0-inverse suffix remain intact.

PROOF.md gives the ordered operator identity and exact ownership/lifetime bill. WORD.json contains the complete 3,935-event replacement word. The verifier authenticates 22 canonical inputs before reusing the credited source algebra, replays the whole word on both original and independent-current routed symbols, checks all 73 frame coordinates at each ADD, and verifies both signed native kernels on 114,688 basis columns with all 224 occurrences bound to their recipes. Eight semantic failure controls cover native offset, maps, sign, ownership, dirty initialization and zero return.

Validation: normal and optimized read-only replay; Git-free export with no executable search path; source corruption rejected before inherited execution; full-word and type-sensitive receipt mutation controls; generation refuses overwrite. Independent AI code/proof review by the coordinating agent and a separate proof/native review found no blocker; their exact scope and file hashes are recorded. This is substantial OpenAI Codex-assisted work, not independent human or formal-kernel review.

This is a local same-output improvement, not a complete physical supplier. Storage, movement, precise outer invocation and productive-family coverage, changed numerical defects, and both all-depth supplier caps remain open. The old one-consumer-per-pair fixed-coverage screen rejects 224 H1, but that rate is hypothetical and differs from the actual first-cut frequency; this is explicitly not a global rejection or a 224-H1 lower bound. The separate incidence contribution is not an executable dependency.

Canonical main: 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority revision 6. Assigned parallel follow-up of #3963; shared claim and progress. Main was fetched and the binding baseline rechecked before publication. No frontier or selected science changes.

Predecessor credit: ywz65 / #2707 for the full F17 decoder and native normalization; Arham Tahir / #2681 for ordered inverse algebra; retained QQ16/native/geometry lineage; #3952 and #3963 for the motivating demand and boundary audits; Nick Petalas / #3339 for the conditional consumer-budget accounting boundary. Exact source heads and adapted code are recorded in PREDECESSORS.json and SOURCES.json.

One bounded follow-up was rejected before publication: replacing the 104 corrected W query copies by borrowed current Y gives a nominal 120-H1 word, but the full routed replay fails at Z0. All 104 W ports have already received inverse-prefix corrections (the eight T ports also receive Z); the bypass loses their future-query contribution. The 120-H1 nominal bill is not an implementation. The published 224-H1 word and receipt remain unchanged.

Registered local replay also passed: PASS_REGISTERED_RESEARCH_PACKAGE outcome=completed. After replay, one extra EOF blank line in budget.py was removed with parsed-AST equality checked; the complete scientific word and receipt are unchanged. Final byte/source-history preflight is recorded below.

Final published head: 400df7fca8f60c19bf163c77564fe3fbe9154b04, created through GitHub. PASS_RESEARCH_COMMIT_BYTES checked all 37 package/input paths against this head and confirmed canonical source-history ancestry; the base-to-head whitespace check passed. Final metadata/source-pin validation passed after the EOF-only cleanup.

Hosted verification passed in run 39219936311 for head 400df7fca8f60c19bf163c77564fe3fbe9154b04. Completed logs confirm the synthetic merge cba74ef20f825a22eb229daaea2f9440b6259999 into canonical base 20ecf2891d80b5acf18d980b2164fb44ea7faa90, CPython 3.12.15, the complete primary verifier, normal/optimized and Git-free replay, all word/receipt/source/overwrite controls, and PASS_REGISTERED_RESEARCH_PACKAGE. The final committed-input check passed with no diff. Workflow conclusion: success at 08:09:05 UTC. This verifies the registered local contract and evidence; physical hosting, numerical-error and all-depth supplier obligations remain open.

[Clear1nsight]

CLAIM / bounded expansion decision — four contrasting QQ16 successor tests; expires 2026-10-19 09:26 UTC.

Canonical main 20ecf28 / priority revision 6 fetched and guidance reread at round start; selected κ remains 836330098538459/500000000000000000. Predecessor #3985 head400df7fca8f60c19bf163c77564fe3fbe9154b04 has author/coordinating-AI replay plus successful hosted registered replay, not human/kernel admission. #3976 and #3977 are likewise checked local scope. Prior completed claim/results: #3963 comment6262031944.

Boundary: all224 original QQ16 X/Y outputs in F17 from224 borrowed current fields, unchanged borrowed values/frames, zero-return owned workspace, owner-private scratch cleared before native calls. Actual gauge44 supplier mapping and arbitrary productive-family allocation are still missing. This round asks which temporary/source obligations can disappear at that exact local boundary, not whether a free decoder or physical supplier exists.

Four separately owned mechanisms/packages, each with a separate focused PR only if useful:

1. Graph pebbling/register allocation: derive exact output read sets and deferrable output initializations; test hosting the rectangular query/response lifetimes in already-owned output slots, including reuse at last query use.
2. Streaming linear algebra/external-memory scheduling: eight persistent full-cover owned F16 query accumulators plus one reusable conjugation buffer in place of112 simultaneous copies. Explicitly price scalar/clear/native/map events and expose traversal/control obligations.
3. Reversible computation/dirty ancilla: exact before/after commutator on borrowed dirty buffers with content/frame restoration; compare the strict source-preserving version and one bounded aligned-owned-source pivot against existing2707 rather than call a dominated reversion a saving.
4. Representation theory/noncommutative harmonic analysis: derive the actual lost inverse-prefix correction in the failed nominal120-H1 shortcut; test one structured routed Schur factorization, with owner-order constraints and a literal paid word or a scoped obstruction.

Ledger: predecessor224H1/448wrappers/352maps/27184scalar and112extraowned fields are constructively upper-bounded for its local exact word. New event/ownership bills will be proved or constructively upper-bounded by literal replay; physical traversal, global hosting/coverage, numerical precision, opposite orientation and both all-depth supplier caps remain UNKNOWN and are not zero. Conditional553-H1 module headroom from3976 is only an optimistic screen, not an admitted budget.

Decision: narrow expansion to these distinct lifetime/contract/source-saving tests for one hour, reserving finalquarter for verification. Each must name real eliminated work and show it does not reappear at consumer/cleanup. Stop after one bounded pivot if exact replay fails, favorable complete screen fails, or the construction is dominated with no changed obligation; preserve a discriminating fixture/handoff instead of expanding a census. A κ improvement requires the missing supplier mapping and complete global bill, not local counts. No selected frontier change, merge or closure.

Overlap: repository helper and targeted live open-PR searches for QQ16 streaming/dirty/Schur/pebbling, current Discussions/start-here/newest handoffs checked; broader remote Discussion/closed-diff coverage remains partial. Distinct predecessors2794/2722 fixed streaming scratch,2651/2617 early output capture,1969 Schur centre transfer,3422 dirty helper and3339 existing count/budget lineage will be credited where used. New local data/proofs remain independent of siblings. Clear1nsight with substantial OpenAI Codex assistance from four research agents and coordinating review; Apache-2.0.