Tooling: exact owner footprint masks reduce checker wall time by 13–19%

[TMM2009]

PR626's independent owner checker builds Python sets for 35,424 helper footprints. This optional, source-pinned adapter represents the same occupied cells with integer bit masks while executing every original check and all exact moment arithmetic. Status: tooling improvement; no new κ.

Final alternating 31-pair local runs of the complete checker measured:

macOS arm64	Original median	Adapter median	Less wall time
Python 3.14.6	0.5620 s	0.4571 s	18.7%
Python 3.12.11	0.5108 s	0.4447 s	12.9%

Every timed receipt matches byte for byte, including exact moment fractions. The reference is the unchanged direct CLI; the candidate pays its source-hash and substitution overhead. Processes are fresh, OS caches are not flushed, and other system activity is uncontrolled. Raw samples, earlier runs, code hashes and the finite-set equivalence argument are included. No full-frontier or multiplication-runtime speedup is claimed.

Reproduce offline with Python 3.11+ and the standard library:

python3 -I -B research/owner-footprint-bitsets/verify.py
python3 -I -B research/owner-footprint-bitsets/test_masks.py
python3 -I -B research/owner-footprint-bitsets/benchmark.py --pairs 31 --out /tmp/owner-mask-benchmark.json

Use a fresh benchmark output path. Expected: PASS_OWNER_FOOTPRINT_MASKS baseline_runs=6 negative_controls=19 positive_controls=1, five tests OK, and PASS_OWNER_FOOTPRINT_BENCHMARK with fresh timings.

Validation includes 18,943 exhaustive small-domain comparisons, 2,000 seed-91 larger cases, sparse boundaries, five caught implementation mutations, and 19 complete-check rejection controls. Repinning lets overlap, gaps, capacity, native-orientation and both failed moment thresholds reach their actual checks. Failure diagnostics agree exactly. Source drift is rejected; a valid row permutation preserves the complete receipt. Normal/optimized baselines, two Python versions, and a minimal relocated package with only 12 pinned inputs pass. Selected-record binding and staged whitespace checks pass. A scoped Linux workflow is included as workflow.example.yml for maintainer installation. It is not enabled because the publishing credential lacks GitHub workflow scope; no remote CI result is claimed.

Base: 66849e1475c930fd7732b1a0067ffee7e1e95f60. Selected #626: 660de893d27bb7abd1d5b01b1640207528a5c851; #622: f69477f4f6d3c7eb1f740b166881bffed4ffa9f6; #623: e39a615ca601ce5d62abeb24f3fcdb8644ca1005. All execution inputs have SHA-256 and Git-blob pins. Native artifacts are archived original PR622 outputs, not a fresh supplier reconstruction.

OS seed 91, verification-engineering track. Three candidates and the local claim/result are recorded. No open sprint coordination issue was available. The final full-queue refresh covered 539 open PRs through #720, followed by a latest-arrival check through #725 immediately before submission. #721 optimizes reconciliation-memory use, a distinct workload. PR672 optimizes the same checker's exponential arithmetic; this changes only footprint representation and does not import or compose that work. PR667/662/707/709/367/452 address other workloads and are credited.

Only the new research package is added. Existing sources, upstream/, defaults and the selected frontier remain unchanged. Repository-wide make verify, supplier cold replay, full controller/47-inequality assembly and Lean were not rerun. All inherited compiler/common-label/mutable-basis/hosting/residual/precision/tape/analytic obligations remain. This is not formal verification of the multiplication theorem.

Original checker: Chafik Boukhalfa/chafreaky; K-triangle providers and dispatch: Jeremy Kothe/jeremykothe2, with the retained broader lineage in NOTICE.md. Prepared for Agastya Sondhi (@TMM2009) with substantial OpenAI Codex assistance, Apache-2.0. No independent human review claimed.