[Clear1nsight]

Change and mathematical scope

A complete capture of all 2,579,750 contribution keys from the frozen PR5864 qualifier closes the actual-data gap in the earlier synthetic key pilot. Every ordered opaque ID, hit/miss/eviction flag and 80-operation/final prefix agrees with an independent Python dict/deque oracle in all four fixed BCCB key-only samples. The scoped audit passes; its performance nomination fails.

The smaller paired actual-key saving is 0.160740552 wall / 0.160850468 CPU seconds. Adding the earlier integer gross-region anchor, while assigning zero retained integer runtime and zero new bill, gives 4.845453128 / 4.844279183 seconds, below the fixed 5.199267809 / 5.1705 second planning thresholds. This conditional cross-run screen is neither a measured complete saving nor a universal upper bound. The further 15% complete PR5864 target remains unmet. No production integration, κ or selected-frontier change is proposed.

The package retains the complete compressed key corpus, one lossless canonical copy of all ordered outputs, hashes for all five original output sets, exact measured sources, source-only reviews and raw bounded diagnostic receipts. Standalone finite/data/context verifiers authenticate those artifacts without replacing the historical performance gate.

Task and overlap

Base: 19fdb086ea96af9a9f87936a2b888e10af4ed810. Publication head: 1018e3751fce3aacf99a5dbdc05d387fb63b6672. Scope: refined replay-runtime / bounded actual-key headroom audit. The original card and selected record remain pinned historical provenance. The five data inputs were separately committed at f901e992a3846ff0351bfb0caf75b6167788cd74; the official PR5864 archive is the existing declared repository input.

This follows PR5908, which assesses four C opportunities. PR5864 supplies the frozen qualifier; PR5875 is the separate negative score-dispatch pilot. PR5929 has a separate H25 complex-side verification segment result and supplies no percentage to this PR5864 gate.

Adjacent memoization precedents include PR2506 (P104 binary matrix-vector cache) and PR4947 (source-bound exact inverse reuse). They cover different workloads; their author-reported timings were not replayed here. A bounded exact-phrase open-PR query is not exhaustive novelty evidence.

Validation

Linux 64-bit CPython 3.12.3 local checks pass from a Git-free copy with source files 0444 and directories 0555:

• python3.12 -B research/verification-key-headroom/verify.py finite and the same with -O: fresh unchanged C comparator, 82 finite receipts and 50 natural rehashes each, clean compiler logs.
• verify.py data: every one of 2,579,750 keys, all ordered IDs/flags and 32,247 prefixes compared with the independent literal oracle, complete payload hashes and exact counters.
• verify.py context: exact 164-file baseline / 177-file measured-stage reconstruction, original C inversion, reference pins and official PR5864 export equality.

Metadata/pins, separate Astra source review, committed-byte/source-history preflight and the 276-file committed export with all six declared inputs pass. All 252 protected source/data/metadata files equal the tested copy; source manifest SHA-256 is 9b3c05b97f926c2c767ad394a399554cd4034a86526dead247251e0921c697d5. Local receipts are in the package's publication/; VALIDATION.json records the local publication state before hosted execution. Exact-head run40542719102 now passes676 shared tests/22 skips and all four registered package commands on CPython3.12.15. Every key/ID/flag/prefix and exact context passed; tracked/index diffs are clean. Hosted success is scoped correctness/transport evidence, not a new performance sample. Final seven-field handoff.

The original complete capture/replay finished in 42.639824143 wall / 42.54 CPU seconds with sampled process-group RSS 437,825,536 bytes under 180-second / 1-GiB limits. It retained 120 finite fixtures, all 25,984 actual events / 799 coordinates, complete signed witnesses, source bytes/modes and original cache statistics. The key comparator passed 82 finite controls and all four full-output comparisons. This outer diagnostic bill is not a complete performance sample.

No new full capture or timing reroll was performed for publication. Optional capture reproduction is not registered for CI. Original Fraction arithmetic and ordered score operations remain unchanged in the measured capture. Both key-only arms exclude Fraction arithmetic, descriptor extraction, callback validation, score updates and full qualification. Hosted success cannot establish the local performance interpretation.

Credit

Retains inherited P104/source authors and Apache-2.0 notices, Ali Romman and PR5848/5864 predecessor credit, plus PR5908/5875 lineage. PR2506 credits Chafik Boukhalfa, Ali Romman and Rohan Arun; PR4947 is by JoyboyBrian. Prepared for Clear1nsight with substantial OpenAI Codex assistance and Astra xhigh design plus separate source review; the parent alone executed guarded science. No independent correctness, C memory-safety, theorem or κ certification is claimed.

Open next questions

See the package README/TASK questions. A continuation requires a materially different source-pinned mechanism, exact ownership/arithmetic/return/witness contracts and credible wall and CPU headroom after retained work and the complete new bill. More favorable repetitions, GMP provisioning or full integration of this failed nomination are not justified by these receipts.

[Clear1nsight]

Status / UTC / priority: Scoped negative audit published; local checks and exact-head hosted CI PASS. 2026-10-27 17:00 UTC, priority revision 8, canonical main 19fdb086ea96af9a9f87936a2b888e10af4ed810 (complete history refreshed; no guidance change).

Concrete question / premise: Does replay of the complete actual PR5864 key sequence fund the further 15% complete wall AND CPU target where the synthetic key pilot failed?

Established / evidence status: All 2,579,750 keys and every ordered ID, flag and 32,247 prefixes agree with the literal dict/deque oracle; normal/-O finite and exact-context checks pass from a read-only Git-free copy. Separate source review passes. The actual-key-only difference plus the earlier integer gross anchor still fails the optimistic zero-residual/zero-bill screen. This is author-executed evidence, not independent scientific acceptance or a universal bound.

Public artifact / exact head: PR5972, 1018e3751fce3aacf99a5dbdc05d387fb63b6672, research/verification-key-headroom. Metadata, committed bytes/source history and the 276-file export with six declared inputs pass. CI run40542719102 passed676 shared tests/22 skips and all four registered commands on CPython3.12.15; tracked/index diffs are clean. This is scoped correctness/transport evidence, not a new performance sample.

Remaining obligations: Further complete PR5864 15% remains unmet; integration, retained integer work and the complete new bill are unpriced. No κ/frontier, general Fraction/API/GC or C memory-safety certification. PR5929's separate H25 segment percentage cannot count toward this target. Independent review/replay remains open.

Next useful task / stop test / runtime: Independent verify.py finite (normal/-O), data, and context checks; each registered timeout is at most180 seconds, existing Linux64 CPython3.12/compiler/headers. Local data check took43 seconds; this is verification cost, not a speedup. No full capture reroll or failed-nomination integration is requested. Reopening requires a materially different source/ownership/cost proposal that preserves the independent oracle and complete outputs.

Active claim / released: This scoped audit and its publication are DONE/RELEASED. No new scientific nomination is funded. All source agents are terminal; no scientific process is running. Kappa, Work1/2 mathematics, feeds and automations remain paused. No merge or selected-frontier change requested.