Audit: bind retained PR589 supplier identity and COPY deficit

research/retained-bit-supplier-binding/README.md

@@ -0,0 +1,150 @@
+# Bind the retained PR589 supplier identity
+
+**Supplemental source-binding guard and finite audit; no new kappa.**
+
+[PR1277](https://github.com/CrocSwap/integer-mult-bounds/pull/1277) identifies a
+remaining question: the PR799 assembly can name a false `bit_supplier` commit
+and Git blob while retaining its accepted numerical result. Its primary finding
+and suggested ancestry-equality repair belong to Sagar Thakrar. This package
+answers the separate supplier-identity question with an opt-in executable guard.
+It changes no existing file or production acceptance predicate.
+
+All four committed PR741/PR799/PR1042/PR1094 assemblies and the archived PR589
+bit result pass. The guard independently connects the cited source to its
+actual word and deficit, rather than treating a successful endpoint comparison
+as proof of provenance. Normal verification is read-only.
+
+## Exact boundary and evidence
+
+The trusted historical selection is main
+`bce54fb3f895cbe5cd4c0eb477c3afeed100190f`; its reviewed PR589 head is
+`a5afc8c6169ea61ba7c51dfd3cade02cc7b13b46`.
+[SOURCES.json](SOURCES.json) pins every external input and the governing source
+code. The guard imports no supplier, frame checker or moment engine.
+
+1. **Git provenance.** The original source uses `research/e8-bit-p111/`.
+   Main retains it as `research/e8-bit-p111-tuned/`. The EXPECTED record and
+   compressed word are byte-identical at the exact PR589 head, the integration
+   commit `b3eb785ac5b4b1ca2beea0f80db078a505626a90`, and the pinned live inputs.
+   Git confirms that PR589 is a direct parent of that integration commit and
+   that the integration precedes this package's base. The original-to-retained
+   rename is checked explicitly; the assembly's retained path is not assumed to
+   have existed on the original contributor branch.
+2. **Word-to-deficit binding.** Following the source format's D/A frame
+   assignments up to COPY and performing separately implemented rational row
+   elimination gives nine ranks `[7,7,7,7,7,7,7,7,7]`, hence decoder constant 63.
+   With the retained nine-stage, 32-replica layout, `K=4`, `M=64`, `v=111`,
+   `W=28416`, the full recorded profile has rank mass 1808352. Both sides give
+   `MW - mass = T(2Kv - L*cst) = 10272`.
+3. **Consumer guard.** Every assembly must name that exact source commit,
+   integration commit, repository, PR, retained path and EXPECTED blob, and the
+   independently derived word/deficit. Its bit input digest, endpoint, decimal
+   and bad-residue surcharge must agree with that source. A supplied bit result
+   must preserve the entire priced profile, inventory and recorded bit budget.
+
+The EXPECTED Git blob is `e293ddc557742637f9f679f083a1a282d87b1b13`; the
+compressed word blob is `85a4499bf9e4911a705d50567a2c2ef6cf23739a`.
+The decompressed word SHA256 is
+`5642fd0e35a1acc11acff8d0c08350c1617e95f310d4d35ad63a651f9eeeba0b`.
+These are source pins, not hashes of newly regenerated gzip containers.
+
+[RESULT.json](RESULT.json) is deterministic. Ten test groups include 23
+rejection fixtures: six false provenance fields with unchanged endpoint and
+kappa; a different real commit containing the identical EXPECTED blob; a
+compensating `[6,8]` COPY-rank substitution preserving total rank 63; an actual
+COPY-frame change; profile mass/type faults; endpoint/digest/surcharge/deficit
+faults; bit-result faults including a rank-mass-preserving histogram change;
+the CLI zeroed-source fixture; and corrupted relocated source bytes.
+
+The positive tests cover the four actual assemblies, the actual bit result,
+and physically relocated inputs using the same Git object database. Checks run
+under `-I -O -B`; this guard does not rely on Python assertions.
+
+## Reproduce and use
+
+Python 3.12 or later, Git, full repository history and no Python dependencies:
+
+```sh
+python3 scripts/research_package.py --package research/retained-bit-supplier-binding --check-only
+python3.12 scripts/research_package.py --package research/retained-bit-supplier-binding
+```
+
+Expected: `PASS_RETAINED_PR589_SUPPLIER_BINDING assemblies=4 deficit=10272`,
+ten passing tests, and `PASS_REGISTERED_RESEARCH_PACKAGE`.
+
+After a replay that retains **this exact PR589 supplier**, additionally guard
+its actual outputs:
+
+```sh
+python3 -I -O -B research/retained-bit-supplier-binding/verify.py \
+  --assembly /absolute/replay/global/ASSEMBLY.json \
+  --bit-result /absolute/replay/ancestor/bit/RESULT9.json
+```
+
+Paths are explicit; the guard never follows a receipt's machine-local ancestor
+path. A different word or source version requires a new contract and pins.
+`--write /absolute/receipt.json` explicitly writes a freshly checked receipt;
+otherwise the default fixture run also compares against committed RESULT.json.
+
+## What this establishes, and what remains
+
+This adds the executable supplier-identity check requested in PR1277; it does
+not replace PR1277's broader equality-to-the-recorded-ancestry repair. The
+earlier finding that a zeroed supplier passes `check_selected` is credited to
+PR1277 and was not freshly replayed here. This package freshly tests rejection
+by the new guard, including its real CLI. It does not claim that the unchanged
+production pipeline now rejects the fixture.
+
+The frame trace is used only to measure COPY ranks, not to establish frame
+legality, scalar restoration or H1-H5. The recorded global profile and bit
+budget remain inputs. No native/scalar/phase/emitted-controller replay,
+orthogonal-chart proof, cold supplier rebuild, certified moment endpoint,
+complex-side verification, all-size/compiler, hosting, precision/tape,
+exact-recovery or analytic-transfer proof is performed. No full multiplication
+theorem, runtime improvement, selected-frontier change or execution
+authentication is claimed.
+
+## Task and overlap
+
+Card `supplier-boundary-audit`, refined to the PR589 identity boundary; chooser
+nonce `78116c731f9a59be32e8cd1544265921`, adversarial-audit track, priority
+revision 2. [TASK.json](TASK.json) records the scope, method, domain and stop rule.
+The card's historical owner audit and current `research_baseline.py` both pass
+as orientation controls; they are not new results.
+
+An entry snapshot covered 806 open PRs including drafts, with complete PR
+pagination. The repository overlap helper searched their titles/bodies and
+local top-level package notes/ledger. Discussions' start-here, listing and live
+handoff thread were read, but complete Discussion pagination was unavailable.
+The browser was signed out and the GitHub connector exposes no Discussion-write
+tool, so the claim is local. No absence-of-overlap or novelty proof is asserted.
+Final refresh details are recorded in TASK.json and the PR.
+
+Nearest work: PR1277 at `02d279515ff1fc468ab99a19d89461f1b6391507` owns the
+finding and supplies the explicit next question. PR1307 audits the selected
+bit allocation's physical projectors; PR1312 audits complex-owner X2 coverage.
+Neither is an input here. This package adds source-to-Git and word-to-deficit
+enforcement, not another owner-window or projector census.
+
+## Open next questions
+
+Should a future unpinned replay wrapper require both this guard and PR1277's
+ancestry equality? This would protect the cited supplier identity as well as
+the particular retained ancestry receipt. The smallest useful step is to apply
+both in a successor wrapper, check PR1277's false-supplier/timing variants and
+then run a complete cold replay. Maintainer acceptance and versioned source and
+coverage pins block adoption; changing frozen supplements is outside this
+contribution. The structured question is in TASK.json.
+
+## Credit
+
+Sagar Thakrar / PR1277 for the finding and next question; DreamingOfClouds /
+PR589 and eumemic / PR555 for the word and retained checker/hosting lineage;
+PR545's rank-mass identity; Douglas Colkitt for the integration and composition
+bindings; Rohan Arun, Jeremy Kothe, Hao Yan and the selected complex lineage;
+nightangelflowerwin-ops / PR865 and Lucas Steinmetz / PR838 for the replay
+coverage lineage. Existing source notices and licenses are retained.
+
+Prepared by Arham Tahir with substantial OpenAI Codex assistance in research,
+implementation, verification and writing. No subagents or independent human
+peer review. New material is Apache-2.0.

research/retained-bit-supplier-binding/RESULT.json

@@ -0,0 +1,47 @@
+{
+  "status": "PASS_RETAINED_PR589_SUPPLIER_BINDING",
+  "base_commit": "bce54fb3f895cbe5cd4c0eb477c3afeed100190f",
+  "supplier": {
+    "pr": 589,
+    "repository": "CrocSwap/integer-mult-bounds",
+    "commit": "a5afc8c6169ea61ba7c51dfd3cade02cc7b13b46",
+    "merged_as": "b3eb785ac5b4b1ca2beea0f80db078a505626a90",
+    "path": "research/e8-bit-p111-tuned/tune/EXPECTED.json",
+    "git_blob_sha1": "e293ddc557742637f9f679f083a1a282d87b1b13",
+    "word": {
+      "v": 111,
+      "decoder_constant": 63,
+      "copy_ranks": [
+        7,
+        7,
+        7,
+        7,
+        7,
+        7,
+        7,
+        7,
+        7
+      ]
+    },
+    "derived_deficit": 10272,
+    "deficit_formula": "T(2Kv-L*cst), L=9 T=32 K=4 M=64"
+  },
+  "git_objects": {
+    "tune/EXPECTED.json": "e293ddc557742637f9f679f083a1a282d87b1b13",
+    "tune/bitE8mut-p111-tuned.json.gz": "85a4499bf9e4911a705d50567a2c2ef6cf23739a"
+  },
+  "record_sha256": "b9f0e5ca038abe837b0007c8fe006f366a6f02009244ac57705a8a200e2301f4",
+  "word_content_sha256": "5642fd0e35a1acc11acff8d0c08350c1617e95f310d4d35ad63a651f9eeeba0b",
+  "assemblies_checked": 4,
+  "bit_results_checked": 1,
+  "assembly_sha256": [
+    "21e2df9286542c9e2786f3720d0f13de0d0a7a982761ebe2184f8f7f811c774d",
+    "65c7648a8edaeaf4e0b8de3e21eeda94fba124d0935010d45ac9f76375b72cec",
+    "093f35445812ea96fe1bfe6a3af7d769cb7f274e08f4fb30f71a17ed1161e54e",
+    "1122cb3a575f8c788345fc1bf4c108ac7c3250fb81a5988671f5eb04b3ab3045"
+  ],
+  "bit_result_sha256": "28795393651ef4fbbc6232eb2b32f7b39b91fbd5ea6d2315583f350714698055",
+  "live": 28416,
+  "rank_mass": 1808352,
+  "scope": "Pinned PR589 source identity, COPY-frame ranks and profile deficit; no scalar/phase/native replay, certified moment or all-size theorem."
+}

research/retained-bit-supplier-binding/SOURCES.json

@@ -0,0 +1,84 @@
+{
+  "schema_version": 1,
+  "base_commit": "bce54fb3f895cbe5cd4c0eb477c3afeed100190f",
+  "files": [
+    {
+      "path": "certificates/selected-result.json",
+      "usage": "provenance",
+      "git_blob": "ae6a99f28aa468649eb66a361e764d769363bf11",
+      "sha256": "10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "c05fafaa25ba95b2d5af86b1800427198e3d20eb",
+      "sha256": "89b5928bc92c81e2647b6d702a7ad85a6795b8f44e2eb267b2ff3145d8f2b2ce"
+    },
+    {
+      "path": "research/community-frontier1042-audit/ASSEMBLY.json",
+      "usage": "input",
+      "git_blob": "4e308dc23da97373d66e5d8d4a68b0661caea8e9",
+      "sha256": "093f35445812ea96fe1bfe6a3af7d769cb7f274e08f4fb30f71a17ed1161e54e"
+    },
+    {
+      "path": "research/community-frontier1094-audit/ASSEMBLY.json",
+      "usage": "input",
+      "git_blob": "5fd676df060f3b450782fdf5ff5f6beb63a5eeb0",
+      "sha256": "1122cb3a575f8c788345fc1bf4c108ac7c3250fb81a5988671f5eb04b3ab3045"
+    },
+    {
+      "path": "research/community-frontier741-audit/ASSEMBLY.json",
+      "usage": "input",
+      "git_blob": "d2a594a6fdb33907f8e6760aa00fd4650535930f",
+      "sha256": "21e2df9286542c9e2786f3720d0f13de0d0a7a982761ebe2184f8f7f811c774d"
+    },
+    {
+      "path": "research/community-frontier741-audit/bit/RESULT9.json",
+      "usage": "input",
+      "git_blob": "d9f15563217bce0d348a63d46e930d0ae997e211",
+      "sha256": "28795393651ef4fbbc6232eb2b32f7b39b91fbd5ea6d2315583f350714698055"
+    },
+    {
+      "path": "research/community-frontier799-audit/ASSEMBLY.json",
+      "usage": "input",
+      "git_blob": "a04bfd932a15196fe8e2418eae912ea745152593",
+      "sha256": "65c7648a8edaeaf4e0b8de3e21eeda94fba124d0935010d45ac9f76375b72cec"
+    },
+    {
+      "path": "research/e8-bit-p111-tuned/code/xcheck_p.py",
+      "usage": "input",
+      "git_blob": "184f5dcf94a486cc7db91ca4db5865916bf26009",
+      "sha256": "af3decc877e27c00001cf56f50c6c09566e8e101297ab34c7860aed96a06d80e"
+    },
+    {
+      "path": "research/e8-bit-p111-tuned/tune/EXPECTED.json",
+      "usage": "input",
+      "git_blob": "e293ddc557742637f9f679f083a1a282d87b1b13",
+      "sha256": "b9f0e5ca038abe837b0007c8fe006f366a6f02009244ac57705a8a200e2301f4"
+    },
+    {
+      "path": "research/e8-bit-p111-tuned/tune/bitE8mut-p111-tuned.json.gz",
+      "usage": "input",
+      "git_blob": "85a4499bf9e4911a705d50567a2c2ef6cf23739a",
+      "sha256": "e3e301e62bc2b41dbf1ceac0acf3d35824ebeab1848c5a2142d4577b109da52f"
+    },
+    {
+      "path": "research/gauge44-frames-x/code/assemble.py",
+      "usage": "input",
+      "git_blob": "8cee90b50e238bd1519a7aeb7395f62d277b8ae4",
+      "sha256": "4c270a996b380ab712f72b81173ab24f1c73203a33578f261bfbaa7e67d75aec"
+    },
+    {
+      "path": "scripts/verify_frontier741_bindings.py",
+      "usage": "input",
+      "git_blob": "dcb4029aeaafe60f4d852a1975b01e02438595fc",
+      "sha256": "eff455167e2fd0bc9a67d35241d0e69d8edc58f3a23e776dcfe8f500ca8e8d67"
+    },
+    {
+      "path": "scripts/verify_frontier799_bindings.py",
+      "usage": "input",
+      "git_blob": "ff902406c7acc760cb3ab5c9cf732768bdde1620",
+      "sha256": "43897b3c99212d6134870898a1840f32898aa1b3679b299e654901568ad7e19b"
+    }
+  ]
+}