[mihir-s-05]

PR #4441's helper retiming changes the recursive rank distribution. This package independently reconstructs the link from its literal frame contacts to its complete complex invoice, without importing the predecessor's synthesizer or verifiers. No accounting discrepancy was found in the checked scope.

Each full native invocation removes 191 paid calls while increasing both productive and paid rank mass by 133. Across all 19 invocations, the complete controller has 69,273 paid calls (3,629 fewer), 11,560 productive fields, and rank masses 260,798 / 259,998, preserving surplus 800.

Start with the audit README, independent implementation, and derived receipt.

Checked scope

• Binary Gaussian elimination validates frame bases and nested paths for all 632 persistent roles in both orientations. All 117 entrance, 42 exit and 395 scalar-frame patches are bound to their original contacts.
• A separate exact Laurent replay executes both scalar boundaries and all 164 controller events, derives the main diagonal, and retains every full call, query shear and whole idle charge.
• An empty-patch positive control reconstructs the original invoice. Per-role dimension telescoping independently explains the conserved surplus.
• Thirteen rejection controls cover dropped queries/idles, wrong signs and coefficients, infeasible frames, omitted patches, false productive credit and altered source bytes.

This is a finite accounting audit, not independent acceptance of the proposed kappa or full theorem. Native arbitrary-column action, Clifford phases, owned-query/center erasure, physical routing, tails, compiler, BIT, precision, stock and outer admission remain outside this check. No selected record or predecessor file changes.

Sources and overlap

Started from canonical main 2a13eb1827784206202db90e5dc8327cf8a250db, priority revision 7. Audited #4441 head 5f9d871c6bd1b9e34db5a7ea5ef08a908e94db39. Two native inputs are pinned on main; five exact PR snapshots are retained with source paths, Git blobs, SHA256s and credit. Replay requires no unmerged Git objects or network access.

The closing live snapshot covered 302 open PRs and 3,294 Discussion records. Sparse local and closed-PR-body coverage remains incomplete. Concurrent #4446 at 7d05d73f96c27a9d360011e04af69b23fa903757 certifies the same profile and reruns #4441's native code. This contribution adds a separate per-role contact-path reconstruction and conservation check; it claims no competing construction or endpoint. Canonical coordination: Discussion #4443.

Validation

CPython 3.12.0, Windows 11 x64:

python -B scripts/research_package.py --package research/fresh-helper-invoice-audit --check-only
python -B scripts/research_package.py --package research/fresh-helper-invoice-audit
python -B scripts/check_research_commit.py --package research/fresh-helper-invoice-audit

Metadata/pins, normal and optimized registered replays, all 13 controls, committed-byte checks and git diff --check passed. A 15-file export passed integrity checks and both direct modes in a fresh Git-free directory. Scientific code/input commit: 6ffff4e474b97e9fcc1e13164676e372a9d3af80; final documentation/metadata head: 6d535abed7b60d4a99dc54903011a72ec367e901. The full-history ancestry preflight was not run in this shallow checkout. Selected-source baseline orientation was attempted but cannot complete in the sparse checkout; the unrelated selected cold replay was not run. Hosted CI: run 39677439786 stops in the shared planning job: 318 tests run, four failures in tests/test_research_current_status.py, 11 skipped. The package replay is therefore skipped. The four failures concern the selected decimal formatting, historical banner/heading and stale CI-navigation command. They reproduce locally (four failures among six focused tests). Git blob comparison confirms the test, both affected documents and selected record are byte-identical to canonical main 2a13eb1827784206202db90e5dc8327cf8a250db. This PR changes only its new audit directory. The package-specific registered and Git-free checks above pass; no hosted package pass is claimed.

Construction credit stays with @1337isnot1337, Cinna Davis, zxv, Jacob Alan Sussman and the retained native/Clifford/BIT/outer contributors. This audit was prepared for Mihir Srivastava (@mihir-s-05) with substantial OpenAI Codex assistance; no subagents. Apache-2.0.