Research audit: resolve R/R0 revert expectations with exact PR626 accounting

research/r492-parent-accounting-audit/CLAIM.json

@@ -0,0 +1,25 @@
+{
+  "track": "adversarial mathematical audit",
+  "assignment_seed": 70,
+  "randomness": "Python secrets.randbelow(100), backed by operating-system randomness",
+  "base_commit": "66849e1475c930fd7732b1a0067ffee7e1e95f60",
+  "claimed_at_utc": "2026-10-13T04:46:00Z",
+  "expected_finish_utc": "2026-10-13T05:19:00Z",
+  "coordination": "GitHub pinnedIssues returned no pinned issue; claim recorded locally.",
+  "candidates": [
+    "Remapped width-8 spot-helper obligations",
+    "Provider R=492 versus fixed parent R0=504 in invoice accounting",
+    "Assembly receipt admission and exact finite guards"
+  ],
+  "selection": "70 modulo 3 selects candidate index 1; parent/provider accounting is distinct from open #675's timing and bridge audit.",
+  "object": "The selected PR626 invoice's R/R0 accounting and the corresponding subset of PROOF-NOTE-R.md revert claims",
+  "method": "Independent integer recount from pinned receipts, plus isolated one-site mutations of the accounting statements",
+  "deliverable": "Scoped reproducible audit, exact evidence and negative controls; no new kappa",
+  "stopping_rule": "One selected invoice and a fixed, documented accounting-mutation set; stop after reproducibility and source review",
+  "overlap_check": "Open PR metadata through #712 checked; no R0/generalization/revert-accounting duplicate found. #675 concerns owner timing; #704 concerns review-ledger metadata.",
+  "status": "completed research audit",
+  "result": "Independent selected-invoice recount agrees exactly. The R0 introductory pass expectation contradicts its own table and evidence; implementation rejects the wrong substitutions. Seven source mutations and four altered inputs reject; one equivalent control passes identically.",
+  "pull_request": "https://github.com/CrocSwap/integer-mult-bounds/pull/724",
+  "final_overlap_check": "Queue refreshed through #722 and pinnedIssues rechecked. #717 prices moment sensitivity, not this finite R/R0 invoice or its revert expectations. origin/main remains the exact recorded base.",
+  "verification": "Normal and optimized Python; deterministic EXPECTED equality; clean committed-file export in unrelated cwd; modified source rejection; make selected-record-check. No full cold replay or full make verify."
+}

research/r492-parent-accounting-audit/EXPECTED.json

@@ -0,0 +1,122 @@
+{
+  "exact": {
+    "R": 492,
+    "R0": 504,
+    "adapters": 437040,
+    "charges": {
+      "bridges": 16128,
+      "data_sweeps": 32256,
+      "helper_routes": 70848,
+      "moves": 7168,
+      "parent_banks": 4536,
+      "parent_idle": 16128,
+      "parent_live": 11704,
+      "parent_physical": 12218,
+      "shifts": 3584,
+      "source_sweeps": 16128,
+      "wrappers": 4713984
+    },
+    "compiler": 25934336,
+    "conservative_pervertex": 4711738180507680,
+    "nonlocal_items": 4904682,
+    "outside": 39242128,
+    "parent_live": 11704,
+    "parent_physical": 12218,
+    "pervertex": 1833791190783112,
+    "rebase_delta": -300156,
+    "rebased": 1833791190482956
+  },
+  "finding": "The introductory R0 pass expectation contradicts the table and evidence; the accounting slice rejects R0-to-R substitutions.",
+  "input_controls": [
+    {
+      "delta": 12,
+      "field": "physical",
+      "rejection": "physical stock"
+    },
+    {
+      "delta": -144,
+      "field": "wrapper_elementary_factors_charged",
+      "rejection": "wrapper recount"
+    },
+    {
+      "delta": -1,
+      "field": "selected_real_component_primitives",
+      "rejection": "scalar recount"
+    },
+    {
+      "delta": 192,
+      "field": "local_adapters",
+      "rejection": "compiler and adapter recount"
+    }
+  ],
+  "pervertex_or_rebase_deltas": {
+    "omit_live_rebase": 4536,
+    "parent_banks_use_provider_R": -22407266304,
+    "rebase_64_instead_of_65": 4548,
+    "routes_use_parent_R0": 358516260864
+  },
+  "source_mutations": [
+    {
+      "assertion": "assert invoice['unhosted_pervertex']==pervertex and invoice['hosted_rebased_pervertex']==rebased",
+      "name": "routes_use_parent_R0",
+      "outcome": "rejected",
+      "source_line": 126
+    },
+    {
+      "assertion": "assert invoice['unhosted_pervertex']==pervertex and invoice['hosted_rebased_pervertex']==rebased",
+      "name": "parent_banks_use_provider_R",
+      "outcome": "rejected",
+      "source_line": 126
+    },
+    {
+      "assertion": "parent_banks,rem=divmod(parent_live-2*112*4*8,9);assert rem==0 and parent_physical==parent_live+parent_banks+8+2   # parent R0=504 dedicated helpers per stage bank",
+      "name": "parent_stock_uses_provider_R",
+      "outcome": "rejected",
+      "source_line": 119
+    },
+    {
+      "name": "parent_banks_literal_504",
+      "outcome": "accepted_identical"
+    },
+    {
+      "assertion": "assert invoice['unhosted_pervertex']==pervertex and invoice['hosted_rebased_pervertex']==rebased",
+      "name": "omit_helper_routes",
+      "outcome": "rejected",
+      "source_line": 126
+    },
+    {
+      "assertion": "assert invoice['unhosted_pervertex']==pervertex and invoice['hosted_rebased_pervertex']==rebased",
+      "name": "omit_wrapper_charge",
+      "outcome": "rejected",
+      "source_line": 126
+    },
+    {
+      "assertion": "assert invoice['unhosted_pervertex']==pervertex and invoice['hosted_rebased_pervertex']==rebased",
+      "name": "rebase_64_instead_of_65",
+      "outcome": "rejected",
+      "source_line": 126
+    },
+    {
+      "assertion": "assert invoice['unhosted_pervertex']==pervertex and invoice['hosted_rebased_pervertex']==rebased",
+      "name": "omit_live_rebase",
+      "outcome": "rejected",
+      "source_line": 126
+    }
+  ],
+  "source_slice": {
+    "first_line": 118,
+    "last_line": 153,
+    "strict_checks": {
+      "halving": true,
+      "literal": true,
+      "local_primitive_bound": true,
+      "logical": true,
+      "mass": true,
+      "precision": true,
+      "router": true,
+      "row": true,
+      "tape": true
+    }
+  },
+  "status": "PASS_R492_PARENT_ACCOUNTING_AUDIT"
+}

research/r492-parent-accounting-audit/README.md

@@ -0,0 +1,151 @@
+# Audit of provider and parent helper accounting
+
+The selected PR626 invoice correctly uses provider `R=492` for helper routes
+and fixed parent `R0=504` for its retained bank and stock envelope. This audit
+reproduces that finite accounting and resolves a contradiction in the reported
+revert-control expectations. It does not claim a new kappa.
+
+The precise discrepancy is in
+[`PROOF-NOTE-R.md`, line 26](../complex-k492-owner-repair/PROOF-NOTE-R.md#L26).
+Its introductory expectations say that the R0 controls pass when R differs
+from 504. The table's `emit.parent_R0` and `invoice.parent_R0` rows instead
+substitute provider R into parent stock and report failure. The evidence
+paragraph agrees with those rows. The correct distinction is:
+
+- Substituting 492 for a required parent count of 504 must fail.
+- Substituting the equal literal 504 for the derived parent bank count must
+  preserve the result.
+
+The original pinned note is retained unchanged. The
+[maintainer review](../../docs/research/community-k492-review.md) already says
+the scratch 32-site revert suite was not submitted. This package supplies new
+evidence for a smaller, explicitly bounded accounting audit. It does not
+reproduce that missing suite or its R=504 fixture.
+
+## Reproduce
+
+From the repository root, using only the Python standard library:
+
+```sh
+python3 -I -B research/r492-parent-accounting-audit/verify.py
+```
+
+Expected output:
+
+```text
+PASS_R492_PARENT_ACCOUNTING_AUDIT: 9 finite guards, 7 source rejections, 1 identical control, 4 input rejections
+```
+
+`--emit` prints the deterministic evidence in [EXPECTED.json](EXPECTED.json).
+The default command also checks that evidence for exact equality. The command
+works from another working directory when given the script's absolute path.
+It also works under `-O`: the audit's own checks use explicit exceptions, and
+the selected source assertions are compiled with optimization disabled.
+
+Validation passed with normal and optimized Python, and from a committed-file
+export containing only this package and the nine pinned inputs in an unrelated
+working directory. Normal and optimized evidence is byte-identical. Changing
+a pinned source byte is rejected before replay. `make selected-record-check`
+also passes. The full `make verify` and cold supplier reconstruction were not
+rerun because no supplier, shared verifier or selected artifact changes here.
+
+## Exact result
+
+The independent recount reads the parent and provider word dimensions, the
+selected dispatch and the per-provider, per-orientation invoice entries.
+It does not import the supplier's arithmetic engines. It retains the parent's
+nine-bank geometry and the published scalar primitive and wrapper receipts
+as assumptions, then sums their charges independently.
+
+| Quantity | Exact value |
+|---|---:|
+| Parent helper count, from 728 columns minus 224 data columns | 504 |
+| Provider helper count, from 716 columns minus 224 data columns | 492 |
+| Helper routes, two directions across 72 complete calls | 70,848 |
+| Parent bank charge, nine banks of 504 helpers | 4,536 |
+| Parent live / physical stock | 11,704 / 12,218 |
+| Hosted live / physical stock | 7,168 / 7,670 |
+| Compiler factor | 25,934,336 |
+| Nonlocal items / outside work | 4,904,682 / 39,242,128 |
+| Complete parent-envelope price P | 1,833,791,190,783,112 |
+| Rebase correction | -300,156 |
+| Hosted rebased price | 1,833,791,190,482,956 |
+| Retained prior allowance plus complete fresh P | 4,711,738,180,507,680 |
+
+The stock equations are
+`parent_live = 7168 + 9*504`,
+`parent_physical = parent_live + 504 + 8 + 2`, and
+`physical = 7168 + 492 + 8 + 2`.
+The rebase correction is
+`65*(7670-12218) + (7168-11704) = -300156`.
+All charge categories are recorded individually in the evidence.
+
+The audit also extracts the original `assemble.py` statements from the
+`parent_live` assignment through the last finite-invoice check, lines 118–153
+at the pinned base. It executes these unchanged statements on the selected
+receipts and checks agreement with the independent recount. The extraction
+has unique AST boundaries and the complete source file is hash-pinned.
+The nine finite guards use the original implementation; they are replayed,
+not independently reimplemented. The invoice recount is the independent part.
+
+## Controls and their meaning
+
+The eight source controls each change one uniquely matched expression in an
+in-memory copy. Seven fail at the parent identity or exact invoice comparison.
+Replacing `9*parent_banks` by `9*504` preserves every captured result. Four
+additional input controls exercise the independent recount by changing stock,
+wrapper work, scalar work or local adapter work. No source file is modified.
+
+With all other inputs held fixed, changing provider route R to 504 increases
+P by **358,516,260,864**. Changing only the parent bank R0 to 492 decreases P by
+**22,407,266,304**. Replacing the rebase coefficient 65 with 64 increases the
+rebased price by **4,548**. The source checks detect each mismatch.
+
+The route substitution overcharges. Its rejection establishes disagreement
+with this exact invoice, not that every enlarged conservative allowance is
+mathematically unsound. The controls are concrete failed checks. They are
+neither counterexamples to the selected construction nor impossibility results.
+
+## Pins, scope and credit
+
+Base commit: `66849e1475c930fd7732b1a0067ffee7e1e95f60`.
+Assignment seed: **70**, drawn with `secrets.randbelow(100)` from OS randomness.
+[CLAIM.json](CLAIM.json) records the candidate selection, stopping rule and
+overlap check. No pinned sprint coordination issue was available.
+
+The selected source is Chafik Boukhalfa's PR626,
+`660de893d27bb7abd1d5b01b1640207528a5c851`, building on Jeremy Kothe's
+PR622 `f69477f4f6d3c7eb1f740b166881bffed4ffa9f6` and PR623
+`e39a615ca601ce5d62abeb24f3fcdb8644ca1005`. The retained bit supplier is eumemic's
+PR543 `2065fe4f72a914d907f1046603f5a36f17018535`; this audit does not replay it.
+The accepted conditional kappa remains `51110230636731/31250000000000000`.
+[SOURCES.json](SOURCES.json) pins every consumed fixture and source by SHA-256.
+Compressed hashes bind committed inputs; no regenerated gzip bytes are used
+as a test of mathematical equality.
+
+GlobalR and AtomReviewer supplied and corrected the R/R0 generalization.
+The retained envelope and invoice descend from icekylinx's PR531 and the
+PR552/PR558 lineage, with DreamingOfClouds' cold helpers and Rohan Arun's
+recurrence work retained. Their [package notice](../complex-k492-owner-repair/NOTICE.md)
+and [parent notice](../direct-delivery504-global/NOTICE.md) preserve full credit.
+The concurrent [PR675](https://github.com/CrocSwap/integer-mult-bounds/pull/675)
+audits owner timing and bridge omission. It is related context, not a dependency
+or the same accounting experiment.
+The refreshed queue also includes [PR717](https://github.com/CrocSwap/integer-mult-bounds/pull/717),
+which prices moment sensitivity to structural changes. This audit instead
+checks the selected finite invoice and its R/R0 mutation expectations.
+
+Only the named accounting slice is replayed. Earlier scalar, frame, phase,
+whole-owner and wrapper admission remains inherited from the selected review.
+The retained prior allowance is a pinned receipt value, not freshly derived
+here. Moment endpoints and the 47 outer assembly constraints are outside this
+audit. All-size compilation, common-label and mutable-basis realization,
+hosting, normalized residuals, precision/tape and analytic transfer remain
+conditional. Finite arithmetic and rejection controls do not prove the full
+multiplication theorem.
+
+Prepared for **thakrarsagar** with substantial **OpenAI Codex** assistance,
+including a separate agent's arithmetic and code review. This is not
+independent human peer review. New work uses the repository's Apache-2.0
+license. The selected frontier, pinned historical sources and `upstream/`
+are unchanged.

research/r492-parent-accounting-audit/SOURCES.json

@@ -0,0 +1,11 @@
+{
+  "research/complex-k492-owner-repair/code/assemble.py": "1b1a9024ac910dc9b9d9e30150684e1fa9cb228a4c1c275b75ab04b57fdef1dd",
+  "research/complex-k492-owner-repair/code/invoice_allcold.py": "141bd2fd58bbfb5b8d0fc8f56f61c45586d5608a5af3487c6b883dd987f998ec",
+  "research/complex-k492-owner-repair/PROOF-NOTE-R.md": "22312bb347bdf159e6a01a45c47675e6170d33b258e2fd9ed52aee88fb92c0ed",
+  "research/complex-k492-owner-repair/data/SPEC.json": "2a2577ac526535826fed9d5f7f83f6e2e0ec058a146c64f7286720b995fa2323",
+  "research/community-k492-audit/pr626/replay/mixed-global/audit/PREFIX-AND-INVOICE.json": "367a73f744ea9b7dd6cca5fa3fda47175f560aa3d37e708c9fa1294dbd4b56eb",
+  "research/community-k492-audit/pr626/replay/mixed-global/audit/OWNERS-AND-PROFILE.json": "ec7bc4157e7ab1707bddf99994149c932b51365de7e742508fd182051ca57fc1",
+  "research/community-k492-audit/pr626/replay/mixed-global/base/split-centres/CANONICAL-WORD.json.gz": "66d3c3c70983224650643a0a05a21ac6c6959f84d0f4528650601016f5a76844",
+  "research/community-k492-audit/pr626/replay/mixed-global/base/providers/ck/CANONICAL-WORD.json.gz": "ae3018c0646be06f47d345bb8baf0e6ea9220096a949fa30a8f7656aa4ad6f45",
+  "research/community-k492-audit/pr626/replay/mixed-global/base/providers/jp8/CANONICAL-WORD.json.gz": "7a3fec1022d3b2b1bd986302a327735ec521d17b043f4a5acb96253a56f6193c"
+}