Research: independently bind QQ16 consumer budget to the first F0 caller

[Clear1nsight]

Result and scope

Independently confirms the fixed-profile obstruction in #3908 at head 0d8880e38ed7c88d70eec6a48e5aa829d0fb22f3 and binds it to the first F0 caller audited in #3932.

The raw producer programs and literal #2707 decoder word reproduce the profiles and 336 decoder calls per pair. A new rational midpoint/trapezoid logarithm enclosure gives an optimistic moment in [1.000097677678, 1.000097678284], above one at the selected saving, even after all eight reported rank-eight children are waived. F0 event 9894 is exactly the L546 child; its prefix hash agrees with #3932. All 37 native cases receive a separate profile census.

The new decision evidence is a decoder/centre tradeoff:

• Under centre 64, at most 121 decoder rank-one calls per pair pass the favorable relaxation with rank-eight work waived, or 61 with it retained.
• A hypothetical centre 63 with all 336 decoder calls and rank-eight work retained passes only the optimistic screen, leaving about 0.0000126126 normalized allowance. No changed coverage construction or physical tariff follows.

Pause expansion of the unchanged recipe as a route to the selected target. The next bounded task is to pin an actual F17-native consumer demand, compose it with the literal decoder, identify real eliminated work and derive a changed complete invoice. Scalar/axis speed alone does not delete a recursive charge; a proved all-depth direct-tape replacement is a separately priced changed premise. No global QQ16 barrier, physical caller admission, complete supplier or new kappa is claimed.

Contribution and workflow

Only research/qq16-consumer-budget-independent/ is added: 11 files. Canonical main 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority revision 6, was refreshed during the session; instruction/priority/catalogue/selected/family guidance is unchanged. This is an explicit assignment. Targeted live PR and Discussion overlap was checked; the repository helper lacked gh, so remote coverage is partial.

The original seven-file #3908 package is retained byte-identically in an archive for actual offline replay, with hashes and attribution. Standalone scripts need no unmerged Git objects. Frozen inputs, shared verifiers and selected science are unchanged.

Publication head: 8dec5c663ace6285aeee49ec3f8eeae9a8bc9a58.

Verification

python3 -I -Werror -B research/qq16-consumer-budget-independent/verify.py
python3 -I -O -Werror -B research/qq16-consumer-budget-independent/test_verify.py
python3 -B scripts/research_package.py --package research/qq16-consumer-budget-independent

Local CPython 3.12.3 normal/optimized warning-strict checks pass: eleven behavioral groups, exact predecessor replay, source/semantic/receipt failures and a read-only Git-free export. Registered replay and metadata checks pass. Published-byte plus source-history preflight passes at the head above for 18 files; diff whitespace checks pass. The original #3908 registered replay and previously unrun source-history preflight also pass at its exact head (12 files). VERIFICATION.json records actual runs; hosted checks are separate.

Full inherited native/bridge/decoder algebra, huge payload execution, selected cold replay, Lean, physical invoice and independent human review remain unrun.

Credit and next question

Credit 1337isnot1337/#3908 for the obstruction; dcharb78/#3181 for its profile/cost screen; ywz65/#2707; Ali Romman and the conserved QQ16/native lineage; Douglas Colkitt and coverage-ceiling authors; Sagar Thakrar and the coverage identity lineage; Nick Petalas and retained source notices.

Chris R / Clear1nsight with substantial OpenAI assistance. Apache-2.0. This adds independent checking and thresholds, not the predecessor theorem. README.md, PROOF.md and TASK.json give the changed-consumer handoff and stopping rule, with adjacent #2391/#2461/#3120/#3873 scopes kept distinct.

[Clear1nsight]

FINDING / scoped DONE — changed-consumer contract test

Published artifact: PR #3952, final head 996ff698d3efd0bdce606e240f3b00e6ab91e7fc; package research/qq16-f17-demand-projection/. Claim released; this comment remains the canonical investigation update.

Actual demand: #2512’s 96 ordinary y1_U arrays in F_U1=F17, at immediate post-B1. The new zero-early-retention word uses all 112 current Y fields, no X reads, 96 owned outputs, sixteen owned full-cover temporaries and eight private scratch fields. It checks exact routed outputs, borrowed restoration, arbitrary owned dirt, every output/ADD phase and zero temporary/private return.

Complete local invoice: 112 H1, 224 native wrappers, 208 cover maps, 560 unit ADDs, 384 quarter ADDs, 144 CLEARs; scalar allowance 9,376. Relative to the complete #2707 decoder, it deletes 224 H1, 448 wrappers, 256 maps and 16,464 scalar allowance by omitting future-logger reversal, F0 inverse and unrelated outputs. This changes the output/read contract; it is not a same-output full-decoder improvement.

The old rank-eight-waived obstruction does not apply to this new demand: its favorable moment is [0.999995759681,0.999995760274]. Keeping the reported eight H8 per call instead gives [1.000023149708,1.000023150305]>1 even with all physical overhead free. With those H8 fixed, at least 51 further pair H1 charges must disappear merely to pass the integer relaxation. With 112 H1 fixed, two H8 across the pair pass and three fail, so at least 14 of the reported 16 pair H8 charges must disappear without compensating costs under unchanged coverage. This does not prove any legal retirement.

Next bounded task: bind the actual eight centre-copy outputs per call to one real continuation of this 96-output demand, including source/destination/frame/last use. Prove retirement or continued need; recompute coverage and charge cleanup. Do not grant the waiver free. Global demand, invocation rate, physical movement/placement, precision/dirty bounds, both supplier caps and all-depth restoration remain open. No global barrier, physical caller admission, supplier or kappa claim.

Evidence: local CPython 3.12.3 registered replay and seven boundary groups pass, including read-only Git-free export and source/word/typed-receipt rejection; nine semantic mutants and 57,344 native basis columns. Final-head byte/source-history preflight passes 36 package/input files; diff whitespace check passes. Hosted checks: run 39210769816, passed planning and package replay at this exact final head; CPython 3.12.15, seven tests in 18.918 seconds, unchanged committed inputs. No independent human or proof-assistant review.

Canonical main 20ecf28 / priority 6 refreshed unchanged before publication. Partial overlap coverage and exact source pins are recorded. Credit ywz65/#2512/#2707, Arham Tahir/#2681, thakrarsagar/#2470, 1337isnot1337/#3908, dcharb78/#3181, #3946 arithmetic and retained QQ16/native/coverage lineage. Chris R / Clear1nsight with substantial OpenAI assistance; Apache-2.0.

Original advance claim

CLAIM — changed-consumer contract test (assigned follow-up; one 60-minute session)

Canonical main 20ecf28, priority revision 6, refreshed at round start. Scope: pin PR #2512’s actual 96 ordinary-port y1_U outputs in F_U=F17 at the immediate post-B1 cut; test a zero-early-retention realization using the current 224 X/Y fields and the literal #2707 inverse. Identify deleted native/scalar/map work, charge owned dirt and temporary cleanup, and screen the complete changed local invoice under #3946/#3908’s fixed coverage law. The existing #2512 consumer assumes eight retained T arrays; this audit does not grant them free.

Success/stop: exact routed output and phase equality, borrowed restoration, private cleanup, full local invoice, and either a viable costed conditional path or a precise fixed-model obstruction. No global lower bound, supplier admission, selected-frontier change or kappa claim. Targeted live PR searches identify #2512/#2492 as demand lineage, #2707 as the zero-retention source and #3946/#3908 as budget predecessors; overlap-helper remote coverage is partial because gh is absent. Further overlap and canonical-main refresh before submission. Canonical investigation/update thread is this comment; handoff hub will receive only a pointer. Chris R / Clear1nsight with substantial OpenAI assistance; credit ywz65, Arham Tahir, 1337isnot1337 and retained QQ16 lineage.