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

research/owner-footprint-bitsets/CLAIM.json

@@ -0,0 +1,59 @@
+{
+  "track": "verification engineering",
+  "assignment_seed": 91,
+  "random_source": "Python secrets.randbelow(100), operating-system randomness",
+  "base_commit": "66849e1475c930fd7732b1a0067ffee7e1e95f60",
+  "claimed_at_utc": "2026-10-13T04:40:00Z",
+  "expected_finish_utc": "2026-10-13T05:16:00Z",
+  "scope": "Exact integer bit masks for occupied-cell overlap, union, cardinality and compact-prefix checks in PR626 extras/check_owner_map.py. No arithmetic-engine or construction changes.",
+  "method": "Same-input paired full-check benchmarks; exhaustive small occupancy states, seed-91 randomized larger domains, and repinned whole-check corruptions.",
+  "deliverable": "Isolated source-pinned opt-in adapter, equivalence argument, negative controls and raw timings. No new kappa.",
+  "candidates": [
+    {
+      "task": "Selected-source inventory hashing",
+      "decision": "Not chosen: filesystem-dependent benefit and weaker mathematical focus."
+    },
+    {
+      "task": "PR626 owner footprint set-to-bitmask substitution",
+      "decision": "Chosen: narrow exact finite-set operations, distinct from arithmetic PRs 667/672."
+    },
+    {
+      "task": "R420 closed-frame inclusion by set containment",
+      "decision": "Not chosen: geometry work overlaps the area of PR709; focus elsewhere."
+    }
+  ],
+  "seed_use": "Seed 91 controls randomized occupancy cases, alternating benchmark order, and selection/rotation of owner mutation sites; the representation change is the substantive distinction.",
+  "coordination": "Only open issue found is #650 requesting Discussions; no sprint issue available, so claim is local.",
+  "overlap": {
+    "latest_pr_seen": 709,
+    "relevant_prs": [
+      672,
+      667,
+      662,
+      707,
+      709,
+      452,
+      367,
+      704
+    ],
+    "searches": [
+      "is:open (performance OR tooling OR fast OR reproducibility OR cache)",
+      "is:open \"owner\" \"performance\"",
+      "is:open \"bitset\"",
+      "is:open \"footprint\" \"checker\""
+    ],
+    "finding": "No footprint-mask optimization found. PR672 shares this checker but changes only its exponential arithmetic; neither imported nor composed here."
+  },
+  "result": {
+    "status": "completed tooling improvement; no new kappa",
+    "summary": "Exact footprint masks preserve complete receipts and rejection coverage. Final 31-pair medians reduce full independent-check wall time by 18.7% on Python 3.14.6 and 12.9% on Python 3.12.11.",
+    "reviewable_result": "https://github.com/TMM2009/integer-mult-bounds/tree/codex/owner-footprint-bitsets-seed91/research/owner-footprint-bitsets",
+    "validation": "Five kernel tests, 19 whole-check negative controls, optimized-Python baselines, source pins and minimal relocated replay.",
+    "overlap_refresh": {
+      "open_pr_count": 539,
+      "latest_pr": 720,
+      "conclusion": "No occupied-cell bitmask contribution found. Same-checker PR672 changes exponential arithmetic and is credited, not imported."
+    },
+    "upstream_main_at_refresh": "66849e1475c930fd7732b1a0067ffee7e1e95f60"
+  }
+}

research/owner-footprint-bitsets/NOTICE.md

@@ -0,0 +1,35 @@
+# Attribution and license
+
+This isolated tooling contribution was prepared for **Agastya Sondhi
+(@TMM2009)** with substantial **OpenAI Codex assistance**, including the
+representation change, proof argument, code, tests, benchmarks and writing.
+No independent human review is claimed. Apache-2.0 applies through the
+repository LICENSE.
+
+The original independent owner checker and exact rational interval module
+are **Chafik Boukhalfa / chafreaky's PR626** contribution. The adapter executes
+them from their original repository locations and retains their bytes. The
+original Python-set implementation is the reference oracle; it is not a new
+contribution here. The four replacement anchors derive from that code.
+
+**Jeremy Kothe / jeremykothe2's PR622 and PR623**, with their disclosed Claude
+assistance, supply the K-triangle providers and inherited dispatch/owner
+geometry. PR626 supplies the selected repaired owner map. **GamingPuzzled /
+Joel Pulikkan's PR620** is credited in the maintainer review for the owner
+headroom lead. The frozen native files used here are the archived PR622
+outputs retained by the maintainer review, not newly reconstructed providers.
+
+The cumulative supplier work includes **DreamingOfClouds**, **icekylinx**,
+**Rohan Arun**, **eumemic**, and their credited predecessors. The original
+[package notice](../complex-k492-owner-repair/NOTICE.md), its bundled notices
+and the [maintainer review](../../docs/research/community-k492-review.md)
+remain the authoritative detailed lineage. Their licensing and AI disclosures
+are not replaced by this notice.
+
+Concurrent **PR672** optimizes exponential intervals in the same independent
+checker, and **PR667** optimizes another exact logarithm engine. They informed
+overlap avoidance but are not imported or composed. **PR709** caches R420
+geometry, **PR707** reuses Gram determinant prefixes, **PR662** accelerates
+Clifford helpers, and **PR367/452** optimize other verification workloads.
+This contribution changes only owner occupied-cell representation. No
+priority claim is made for the standard bit-set encoding itself.

research/owner-footprint-bitsets/ORIGINS.json

@@ -0,0 +1,58 @@
+{
+  "base_commit": "66849e1475c930fd7732b1a0067ffee7e1e95f60",
+  "selected_kappa": "51110230636731/31250000000000000",
+  "source_pr626": "660de893d27bb7abd1d5b01b1640207528a5c851",
+  "source_pr622": "f69477f4f6d3c7eb1f740b166881bffed4ffa9f6",
+  "source_pr623": "e39a615ca601ce5d62abeb24f3fcdb8644ca1005",
+  "native_artifacts": "Archived exact PR622 outputs; no fresh supplier reconstruction claimed. Inherited checker requires these immutable compressed bytes.",
+  "files": {
+    "research/complex-k492-owner-repair/extras/check_owner_map.py": {
+      "sha256": "d240458ad39a08c242d136c0f1edbda0cb0b93977897f6543c9f430576c9c7dd",
+      "git_blob": "abf1964e62f3dce04801dd80246626209beb13d7"
+    },
+    "research/complex-k492-owner-repair/extras/exact_intervals.py": {
+      "sha256": "aa3c2327c5d3742248b720a87b2de0bbe2e34ed2af89a578e3c8d5bd2656a9cf",
+      "git_blob": "8a904997c9692fee7c9447e9a0e43748d0021186"
+    },
+    "research/complex-k492-owner-repair/data/SPEC.json": {
+      "sha256": "2a2577ac526535826fed9d5f7f83f6e2e0ec058a146c64f7286720b995fa2323",
+      "git_blob": "cdc4f6ce03e96fd1ffc5a19926e1d0d4bce4e9de"
+    },
+    "research/complex-k492-owner-repair/data/PLAN.json": {
+      "sha256": "8c48ef6ce079916314126c5c3ff0003b32f24cf0fbfb2a033fc93e1e1949b5d5",
+      "git_blob": "434a431937757def4e496301e3d33106f0026fa3"
+    },
+    "research/complex-k492-owner-repair/data/OWNERS.json.gz": {
+      "sha256": "251e82cec1f40e6c1289703667f403104b957f1254dc4f6baa6df9a3d0c751de",
+      "git_blob": "1fadba6f4217269eabc4994b565f3daef3005d64"
+    },
+    "research/complex-k492-owner-repair/comparison/PR623-SPEC.json": {
+      "sha256": "436d3df54e6ce2433d56435afd569708e5fb22bc638571f5134ab12a9f9099f5",
+      "git_blob": "48f6e2a62a121b7f536c432c46ea58575ee0969f"
+    },
+    "research/complex-k492-owner-repair/comparison/PR623-PLAN.json": {
+      "sha256": "c919c366ef1d3528f1884e47dfce858ab3164b099740e4c17cde6f697b3ec458",
+      "git_blob": "e310a7e26f141eadcbdfb9360d60858a2a0a1cd9"
+    },
+    "research/complex-k492-owner-repair/comparison/PR622-NATIVE-HASHES.json": {
+      "sha256": "28479a183387bcb608bd9ee0572f617c17f316b79ce4a6fa1277ed8689fb3431",
+      "git_blob": "05e48488e9d1d5fe98d32d60777ae88cc272c5ab"
+    },
+    "research/community-k492-audit/pr622/replay/mixed-global/base/providers/ck/CANONICAL-WORD.json.gz": {
+      "sha256": "ae3018c0646be06f47d345bb8baf0e6ea9220096a949fa30a8f7656aa4ad6f45",
+      "git_blob": "58cd1be8f28b74bea83c435eee1f52b36a274281"
+    },
+    "research/community-k492-audit/pr622/replay/mixed-global/base/providers/ck/LOCAL-PROGRAMS.json.gz": {
+      "sha256": "b5da54e049cc127a1065190a647cb71cac0c6ba03d9ccdb71719f4fcbfa886f3",
+      "git_blob": "61ef73c1f958e0d26f3e7a3a82bc3826b7098c93"
+    },
+    "research/community-k492-audit/pr622/replay/mixed-global/base/providers/jp8/CANONICAL-WORD.json.gz": {
+      "sha256": "7a3fec1022d3b2b1bd986302a327735ec521d17b043f4a5acb96253a56f6193c",
+      "git_blob": "7fd7cd1fefbb8b0e12397b355d9f6e227813bd58"
+    },
+    "research/community-k492-audit/pr622/replay/mixed-global/base/providers/jp8/LOCAL-PROGRAMS.json.gz": {
+      "sha256": "fa0c196f65bd6ec1424fdf7d789f6972075c7bb02bf43f573ea85eafab31be98",
+      "git_blob": "e22169a1633638213a5c63425a1919c7185f4c43"
+    }
+  }
+}

research/owner-footprint-bitsets/PROOF.md

@@ -0,0 +1,55 @@
+# Exact representation change
+
+The domain is the pinned PR626 independent owner checker on integer-valued
+JSON owner/plan/native inputs that reach its footprint operations. It checks
+`0 <= offset <= offset + w <= cap` before creating the interval. Thus the
+interval endpoints and width used by the substitution are nonnegative.
+Malformed noninteger inputs are outside this equivalence claim; neither
+implementation is presented as a general JSON schema validator.
+
+For a finite set S of nonnegative integers define
+
+    encode(S) = sum(2**i for i in S).
+
+Distinct indices occupy distinct binary digits. Consequently:
+
+- `encode(S) & encode(T)` is zero exactly when S and T are disjoint.
+- `encode(S) | encode(T) = encode(S union T)`.
+- `encode(S).bit_count() = len(S)`.
+- `encode(range(offset, offset + w)) = ((1 << w) - 1) << offset`,
+  including w = 0.
+
+A nonnegative integer m has `m & (m + 1) == 0` exactly when m has a suffix
+of k one-bits and no higher one-bits, including k = 0. Indeed, adding one
+clears the trailing ones and sets the first zero bit. Any higher one-bit
+would survive the AND; if no such bit exists, the AND is zero. Therefore
+this predicate is equivalent to
+
+    S == set(range(len(S))).
+
+Induct over the owner rows. The original dictionary defaults to an empty
+set; the replacement defaults to zero. Assume both represent the same
+occupied cells at a row's start. Every check before interval creation is
+unchanged. The interval formula gives the same new cells; the overlap
+predicate agrees and, on acceptance, union restores the invariant. Both
+stop at the same failed `need` predicate for this row. After all rows, the
+compact-prefix predicate and cardinality agree for every owner, including
+unused owners. Every idle remainder `cap - len(S)` is identical. The paid
+profile, rank-mass check, exact rational moments and outer comparison then
+execute the original code on identical values. Dictionary iteration order
+and all receipt fields are unchanged.
+
+The adapter performs four exact, counted textual substitutions, only after
+checking SHA-256 pins of the original checker and rational interval module.
+It executes every original `need` statement. It neither caches acceptance
+nor changes a moment engine, footprint capacity, window or owner row.
+The same source-only admission is used for both adapter engines. The timed
+reference is the original direct CLI, so the candidate pays its additional
+pin/substitution overhead.
+
+This is a written equivalence argument for the stated finite operations,
+corroborated by bounded tests. It is not proof-assistant formalization,
+a proof of the construction's inherited physical interfaces, or a proof of
+the full multiplication theorem. Tests do not enlarge this mathematical
+scope. The original checker itself is a supplemental audit; its recorded
+success does not replace the native/controller/outer-assembly pipeline.

research/owner-footprint-bitsets/README.md

@@ -0,0 +1,144 @@
+# Exact owner-footprint bitsets
+
+**Tooling improvement; no new κ.** Replace the occupied-cell Python sets in
+PR626's independent owner checker with exact integer bit masks. On the same
+frozen inputs, final local 31-pair runs measured **12.9–18.7% less wall time
+for the complete checker**. Every result receipt is byte-identical.
+
+The scope is one supplemental check of 35,424 helper roles. This does not
+change a supplier, owner map, mathematical bound, selected pointer or default
+verification command. The original source packages and `upstream/` stay intact.
+
+## Reproduce
+
+From the repository root, with Python 3.11+ and its standard library:
+
+```sh
+python3 -I -B research/owner-footprint-bitsets/verify.py
+```
+
+Expected final line:
+
+```text
+PASS_OWNER_FOOTPRINT_MASKS baseline_runs=6 negative_controls=19 positive_controls=1
+```
+
+The command uses the immutable archived PR622 native artifacts already in the
+repository. It copies only its 12 pinned execution inputs into a temporary
+path containing spaces, replays the original direct CLI and both adapter
+engines (including optimized Python), then checks repinned corruptions and a
+valid row permutation. Temporary files are removed; repository inputs are
+unchanged. This is not a fresh reconstruction of the native suppliers.
+
+Additional arithmetic-oracle tests and reproducible timing:
+
+```sh
+python3 -I -B research/owner-footprint-bitsets/test_masks.py
+python3 -I -B -O research/owner-footprint-bitsets/test_masks.py
+python3 -I -B research/owner-footprint-bitsets/benchmark.py --pairs 31 --out /tmp/owner-mask-benchmark.json
+```
+
+Expected: five tests pass in each test invocation, followed by
+`PASS_OWNER_FOOTPRINT_BENCHMARK` for the benchmark. Choose a fresh output path;
+it is never overwritten. Timing has no pass/fail speed threshold. The
+benchmark checks every complete receipt, including the exact moment fractions.
+
+To use the adapter with the pinned checker on a separately prepared work
+directory, pass its original arguments to `adapter.py`, for example
+`--work /path/to/work --out /path/to/new-result.json`. The inherited provider
+defaults are `work/ck-local` and `work/jp8-local`; original `--provider PID=DIR`
+overrides are supported. The default engine is `masks`; `--engine sets` runs
+the reference code through the same adapter. Source drift is rejected.
+Input changes must satisfy the original checker's plan, owner and native
+bindings. The standalone adapter does not claim an audit of a changed supplier.
+
+## What changes, and why coverage is retained
+
+The adapter pins the original checker and rational-interval source, then
+makes four counted substitutions: empty set to zero, interval to bit mask,
+compact-prefix set equality to a binary predicate, and cardinality to
+`int.bit_count()`. Intersection and union continue to use `&` and `|`.
+Every original `need` statement and the complete moment arithmetic execute
+unchanged. The [equivalence argument](PROOF.md) proves these finite-set
+operations agree for nonnegative integer cells, including empty intervals.
+No accepted result or geometry is cached.
+
+Validation covers:
+
+- 18,943 exhaustive subset/interval comparisons for capacities 0 through 8,
+  2,000 seed-91 larger cases through capacity 4,096, and sparse/boundary cases
+  through bit 8,192. The oracle uses Python sets and integer powers.
+- Five actual adapter-source mutations caught by the independent set oracle:
+  omitted offset, off-by-one interval, displaced interval, and two incorrect
+  compact-prefix predicates.
+- Fifteen whole-check mathematical/data controls: overlap, an interior gap,
+  negative offset, capacity overflow, width, duplicate helper, physical family,
+  copy, stage, missing role, paid profile, both failed moment thresholds,
+  native forward/backward disagreement, and changed inherited geometry.
+- Two inherited input-pin and two adapter source-pin controls. Changes are
+  repinned where necessary to reach the stated predicate; the native-program
+  control also updates its immutable-hash declaration intentionally. These are
+  invalid-input controls, not defects alleged against the accepted construction.
+- Exact matching rejection diagnostics for both adapter engines. A shuffled
+  valid role list and the six baseline invocations retain the full original
+  receipt, including idle remainders and both rational moment brackets.
+
+The final full-control receipts agree on Python 3.14.6 and 3.12.11. Normal and
+optimized unit suites pass on 3.14.6; normal unit tests also pass on 3.12.11.
+The selected-record binding check and staged whitespace check pass. A separate
+minimal relocated package replay is recorded in `evidence/SUMMARY.json`.
+A scoped Linux workflow is supplied as `workflow.example.yml` for maintainer
+installation. It is not enabled: the publishing credential lacks GitHub
+`workflow` scope. Local evidence is not a remote CI claim.
+
+## Measured result
+
+All samples start a fresh interpreter. The reference is the **unchanged direct
+CLI**; the candidate includes adapter import, source hashing and transformation.
+The same archived inputs are used for every sample, with alternating order
+starting with the reference (seed 91). OS caches were not flushed, and other
+system activity was not controlled. The raw samples, child CPU times and code
+hashes are retained in `evidence/benchmark-*.json`.
+
+| Interpreter, macOS arm64 | Pairs | Original median | Masks median | Wall reduction | Masks faster |
+| --- | ---: | ---: | ---: | ---: | ---: |
+| CPython 3.14.6 | 31 | 0.5620 s | 0.4571 s | 18.7% | 27/31 |
+| CPython 3.12.11 | 31 | 0.5108 s | 0.4447 s | 12.9% | 28/31 |
+
+An earlier 31-pair 3.14 run measured 21.1%; the seven-pair inline-mask pilot
+measured 14.0%. Both are retained, not substituted for the final measurements.
+The adapter kernel is unchanged between the 31-pair runs. `verify.py` was
+strengthened before the final runs: the physical-family control now reaches
+that exact formula guard, and failure strings must match exactly. The older
+`validation.json` records the earlier, weaker diagnostic check; the two
+`validation-final-*.json` files are the final evidence.
+
+No full-frontier wall-time, memory reduction, practical multiplication speedup
+or universal performance guarantee is claimed.
+
+## Provenance and remaining scope
+
+Base: `66849e1475c930fd7732b1a0067ffee7e1e95f60`.
+Selected conditional κ remains `51110230636731/31250000000000000`, or
+`0.001635527380375392`: #626 with #543, as specified by
+`certificates/selected-result.json` and its actual #626 assembly/review.
+[ORIGINS.json](ORIGINS.json) pins all 12 execution inputs by SHA-256 and Git blob,
+and records the immutable #626/#622/#623 source commits. Native files are
+original archived compressed inputs; the inherited exact-byte native check is
+retained. No claim about regenerated gzip container equality is added.
+
+OS assignment seed **91**, verification-engineering track. Three narrow
+candidates, the local claim, seed use and overlap review are in
+[CLAIM.json](CLAIM.json). No open sprint issue was available. The final refresh
+covered 539 open PRs through #720. PR672 optimizes exponential intervals in the
+same checker; this work changes footprint representation and does not compose
+its kernel. PR667/662/707/709/367/452 address other costs. Their descriptions
+were considered for overlap, not independently verified here.
+
+Repository-wide `make verify`, selected supplier cold replay, full controller
+and 47-inequality assembly, and Lean were not rerun for this isolated opt-in
+adapter. All-size/compiler/common-label/mutable-basis, whole-call hosting,
+residual, precision/tape/exact-recovery and analytic-transfer obligations remain
+inherited. Finite equivalence tests are not a formal proof of the multiplication
+theorem. Original sources and contributor notices are retained; see
+[NOTICE.md](NOTICE.md) for attribution and substantial AI-assistance disclosure.