Research: project a 96-output F17 demand and screen its retained-centre budget

[Clear1nsight]

PR2512 actually demands 96 ordinary y1_U arrays in F_U1=F17. This projects PR2707's complete inverse onto that endpoint, using all 112 current Y fields and no early T retention or current X reads. The changed 1,410-event word returns the 96 owned outputs, clears sixteen owned full-cover temporaries and eight owner-private scratch fields, and leaves every borrowed input untouched.

The complete local invoice is 112 H1, 224 native wrappers, 208 cover maps, 560 unit ADDs, 384 quarter ADDs and 144 CLEARs: 9,376 real scalar primitive allowance. Relative to the full 224-output decoder, this genuinely removes 224 H1, 448 wrappers, 256 maps and 16,464 scalar allowance. It drops future-logger reversal, the F0 inverse and unrelated recovery; it does not lower the tariff of unchanged recursive calls.

The budget conclusion changes in a material way:

• The old rank-eight-waived obstruction does not extend to this reduced demand. Its favorable moment is [0.999995759681, 0.999995760274].
• Keeping the reported eight H8 per call under the same centre 64 pair-rate model instead gives [1.000023149708, 1.000023150305] > 1, even with every physical overhead free.
• With those H8 fixed, the integer screen permits at most 61 decoder H1 per pair, so this word needs at least 51 fewer genuine pair charges merely to pass it.
• With 112 H1 fixed, two H8 across the pair pass while three fail; at least fourteen of the reported sixteen pair H8 charges would have to disappear without compensating costs under unchanged coverage. No legal copy retirement is supplied.

These are conditional fixed-profile necessary-budget results, not a global lower bound or a claim that the eight H8 survive every changed outer demand. The actual published local endpoint is pinned; a full selected outer continuation requiring only those 96 outputs remains unproved. Physical movement/placement, numerical grids and dirty ranges, both supplier caps, output lifetimes and changed-coverage obligations remain open. The favorable passing waiver is not supplier admission or permission to expand construction. No new kappa or frontier change.

Only research/qq16-f17-demand-projection/ is added. PROOF.md gives the ordered identity, physical-frame argument, complete local counts and stopping rule; WORD.json and RESULT.json are checked in full. Compared with PR2512's same logical output and different input/lifetime premises, the new interface needs sixteen additional current Y reads and sixteen full-cover temporaries; it does not claim unconditional dominance or global storage savings.

Validation on Linux CPython 3.12.3, standard library:

python3 -I -Werror -B research/qq16-f17-demand-projection/verify.py
python3 -I -O -Werror -B research/qq16-f17-demand-projection/test_verify.py
python3 -B scripts/research_package.py --package research/qq16-f17-demand-projection

Observed locally: exact demanded routed rows on independent current Y and independent pre-B1 x1/y1 symbols, all 73-coordinate frame phases, 57,344 positive native basis columns, nine semantic mutants, complete primitive invoice and exact rational threshold brackets. Seven boundary groups pass, including a read-only Git-free export, pre-execution source authentication and whole-word/typed receipt rejection. Registered replay and metadata/pins pass. VERIFICATION.json records commands, environment and unrun obligations; published-byte/source-history preflight passes at head 996ff69 for 36 package/input files. Upstream run 39210769816 passed both planning and package replay at final head 996ff69. Its log binds the merge into base 20ecf28, CPython 3.12.15, seven tests in 18.918 seconds, the registered completion marker and unchanged committed inputs. Replay explicitly adapts authenticated predecessor primitives; it is not a code-disjoint reimplementation of all inherited algebra, independent human review, physical tape execution or selected cold replay.

Explicit assigned follow-up of #3946/#3908; canonical main 20ecf28, priority 6, refreshed at round start, budget checkpoint and before publication, unchanged. Frozen sources and selected science are untouched. Targeted live PR searches, visible Discussion handoffs and local semantic-overlap checks found the named adjacent scopes; the overlap helper lacks gh and complete remote snapshots, so coverage is partial. Canonical advance claim/update.

Next bounded audit: trace the actual eight rank-eight centre-copy outputs into one real continuation of this 96-output demand. Pin source/destination/frame/last use and prove legal retirement or continued need, recomputing coverage and charging cleanup. Do not grant the waiver free. TASK.json has concrete first steps, blockers and stop conditions.

Credit ywz65 #2512/#2492/#2468 and #2707; thakrarsagar #2470; Arham Tahir #2681; 1337isnot1337 #3908; dcharb78 #3181; Clear1nsight #3946; Ali Romman, starship-s, Ross Richardson and retained QQ16/native lineage; Douglas Colkitt, Sagar Thakrar and coverage-ceiling contributors. Exact heads and code adaptations in PREDECESSORS.json; original notices remain. New work by Chris R / Clear1nsight with substantial OpenAI assistance. Apache-2.0.

[Clear1nsight]

FINDING / DONE — centre-copy lifetime and actual outer-boundary audit: PR #3963, exact head 82c812c.

The premise needs correction: all eight selected H8 centres are already removed by canonical QQ16 corrected_program, with paid correction580 ADD+16 CLEAR. Independent exact regeneration and712-column scalar replay confirm no second retirement credit. All old C buffers clear before local exit. Existing construction credit stays with Ali Romman/QQ16 and its source lanes.

Pinned the real physical F0/B1 → pair1 idle → bridge → F2 prefix. Eight reachable integer constant-sector collisions have identical96 projected zeros and different bridge outputs. Separately, PR2707 checkpoint2740/next event2741 requires128 fields absent from the projection. These are projection-only/interface obstructions: borrowed current X/Y remain available to a redesigned paid consumer;128 missing fields is not128 new families.

The retained complete fallback costs336H1,672 wrappers,464 maps,scalar25,840 and fails the same conditional pair screen even with H8 absent: [1.000097677880011248,1.000097677880011249]>1. Reduced112H1 projection passes its favorable relaxation; coverage, integer incidence and all physical/precision/cleanup obligations still require proof. Fixed-T centre reallocation is distinguished from literal deletion.

Three user-requested agents audited centre provenance, recovery boundary and arithmetic/applicability, then cross-reviewed; one also reviewed root bridge witnesses/integrated scope. Normal/optimized receipts, registered replay and seven integrity/export groups pass. Publication byte/source-history check passes all28 package/input files. Hosted run 39214876589 passed at exact head82c812c45: CPython3.12.15, seven tests in31.224 seconds, registered completion and unchanged committed inputs. No human/formal review, admitted global supplier or new kappa. Main20ecf289 and priority6 unchanged at three checkpoints; partial overlap coverage disclosed.

Next useful step: one paid composition at this real boundary—recover/encode missing checkpoint information more cheaply than the336H1 fallback, or change the bridge consumer with explicit removed source work. Pin arbitrary-array equations, frames, lifetime/return, integer invocation/coverage and a complete invoice. Do not repeat centre deletion or optimize isolated precision. See PR TASK/PROOF and exact predecessor credits.

Original advance claim

CLAIM — centre-copy lifetime and outer-continuation audit (assigned follow-up, ~60-minute cap)

Base canonical main 20ecf28, priority 6, freshly fetched unchanged; predecessor #3952 head 996ff69. Root pins an actual outer continuation of the 96-output F17 demand; three user-requested agents independently audit copy provenance/lifetimes, a legal replacement or obstruction, and full-budget applicability. Frozen source inputs preserved, no selected-frontier change.

Bounded test: trace all reported eight H8 centre events and their consumers; distinguish selected-supplier charges from actual QQ16 instructions; identify required outer reads and return obligations; establish a literal fully priced retirement schedule or a precise source/interface obstruction. Stop if a required interface or complete allowance is absent. No free centre credit, global barrier or kappa claim. One canonical investigation/update comment here; handoff hub receives a short pointer after completion.

Early local-only evidence under verification: the selected C buffers appear to clear before call exit, and joint-query corrected_program already removes the selected centre phase. PR2707’s actual F0-inverse consumer also requires more than the 96-output interface. These are being independently checked, not yet a new supplier finding.

Overlap: live #3952 head/comments checked; targeted QQ16/centre/lifetime PR search found #3952 and adjacent #2470/#2651/#2895/#2086 scopes. Full remote Discussion/PR coverage is not claimed; helper refresh and final main/head checks follow. Credit retained QQ16/native/decoder/coverage lineage, ywz65/#2512/#2707, Arham Tahir/#2681, 1337isnot1337/#3908 and #3946 arithmetic. Chris R / Clear1nsight with substantial OpenAI assistance and separate AI agents; no independent human review.