[Clear1nsight]

Upstream publication and complete scope

Publishes fork PR #14, preserving exact verified head af86fb28f4fe039d4bb8419850af99a6fc474587. Publication comparison was refreshed against upstream f626db5c3a3d781d8f7695038776c621651f75e1 on 22 October 2026: 24 changed files, 4 contribution commits. The diff includes the scientific transport audit and dedicated workflow, plus the historical-replay/source-role and navigation/CI repairs described below. Shared repair bytes overlap across fork #14–#18 and must be reconciled once; they are not independent scientific contributions. All five research heads have successful exact-head fork workflows. Opening this PR is not a new scientific replay, upstream CI pass, merge or theorem acceptance. Scientific source pins, attribution and scope limitations are retained.

---

Change and mathematical scope

This adds a separately written, source-bound audit of the 423 declared transport/helper-boundary pairs in upstream PR4470 at b9a13c1673e0e9697de410de16e8bbf3207a7363. It compares every entry of each 512 by 512 lowered matrix with the literal Gaussian-dyadic reference operator in both orientations, including absolute phase. NumPy stores exact int64 numerators; explicit powers-of-two denominators and an arithmetic safety cap replace floating-point tolerances. No predecessor compiler or evaluator is imported.

The complete hosted scientific replay now passes. Both workflows are green at publication head af86fb28f4fe039d4bb8419850af99a6fc474587. The registered replay authenticates and freshly executes the unchanged scientific engine, then compares every field with the committed hosted receipt. The result covers 433,152 basis columns and 221,773,824 complex matrix entries.

No new kappa, selected-frontier change, complete supplier admission, or independent human/formal acceptance is claimed. Completeness of the declared pair list, the chronological 632-input endpoint, the actual p16 query/owner protocol, physical costs, precision/storage, and the outer theorem remain separate obligations.

The workflow repair also resolves four stale selected-frontier navigation assertions and the affected historical source-role mismatches. It reuses zxv's exact PR4436 documentation corrections. PR3563's unchanged historical verifier and receipt are replayed against their authenticated original source snapshot rather than falsely applying the old selected-record relationship to PR4359. The locale checker remains a live test of today's document and test module. The D8 package's self-contained scientific verifier remains unchanged; its four historical guidance/selection hashes are retained as provenance. All original invalidation paths and scientific rejection controls remain in place. See the CI repair note and D8 source-role explanation.

Task and overlap

The scientific and fork base remains 2a13eb1827784206202db90e5dc8327cf8a250db, priority revision 7. Canonical main was refreshed to f9f85c1e4fa1e955ff39f10fca447e3e2a4e4adc during this repair and remained unchanged at the final check. No wholesale archive migration or selected-record change is included.

The user-assigned PR4470 native audit refines independent-moment to physical normal forms rather than repeating invoice arithmetic. TASK.json preserves the full original catalogue card. Sources are exported unchanged from exact PR4470 head b9a13c1673e0e9697de410de16e8bbf3207a7363; the missing-span predecessor is PR4447 at 322e0ec764aa7ef7801383feb1f656e460dd8bed. The four consumed exports are pinned to this fork's ancestor commit f4657a0ad4c11a56646c8c6c422f09335b0c73ba, so publication does not require an unmerged foreign commit to be in the runner's object cache.

PR4478's current package already covers arithmetic, compiler parameters, and the outer rows. PR4457 supplies an earlier operator audit, not this changed 423-pair set. PR4464/4466 cover stage-zero commutation and translations. PR4504 independently checks chronological frames and local billing, leaving the physical operators separate. PR4483 advances the separate compiler route. Overlap coverage was targeted live PR/body and package reading, not exhaustive Discussion pagination or a proof of novelty.

The original review was prepared within Clear1nsight's fork. This now submits that verified contribution upstream; no merge or selected-frontier promotion is requested. The next research programme is recorded in the four-agent planning comment; those proposed agents have not been launched.

Validation

Final head af86fb28f4fe039d4bb8419850af99a6fc474587 has:

• Registered research packages run 39717542509, success, including onboarding/planning and all five selected package replays: exact transport, historical-1042-link, locale reads, D8 mixed target, and k8 three-coordinate temporal. Their committed-input preservation checks pass.
• Dedicated exact transport run 39717543211, success, including the complete exact-matrix replay and optimised scientific controls.

The committed RESULT.json is the byte-exact receipt downloaded from the first successful complete replay, run 39712725621, at original head f4657a0ad4c11a56646c8c6c422f09335b0c73ba. Its SHA256 is 53bdd869ebeee0980b139e422043922c87585bef90c2989969532ccb7b289a00. The final registered run reproduces every scientific and scope field. The published scientific verifier remains Git blob 1c9a9d6410e29b7c201547be9f0bef8685c4571b, and all four immutable source exports are unchanged.

Scientific controls include 196 positive checks, seven rejecting controls, a two-column tensor-phase check, and an actual-source phase mutation that preserves rank but fails operator equality. Five new receipt-wrapper tests reject changed scientific/scope fields, missing/extra fields, numeric-type substitutions, corrupt receipts, and changed engine bytes. Six historical-adapter tests cover exact retrieval, live-edit isolation, bad hashes/blob IDs, missing history, Git replacements, unsafe paths, and the original-manifest guard. The five wrapper tests run in normal and optimised modes in the registered workflow.

Use Python 3.12 with numpy==2.2.6:

python -m pip install numpy==2.2.6
python -B scripts/research_package.py --package research/pr4470-exact-transport-audit --check-only
python -B scripts/research_package.py --package research/pr4470-exact-transport-audit
python -I -B research/pr4470-exact-transport-audit/replay.py

The last command supports a Git-free scientific export. Standard package metadata/source authentication and the affected hosted replays passed. The optional full-history publication preflight and a complete selected-frontier cold replay were not run in this repair session. No Lean, Docker, independent human review, or complete physical/analytic theorem audit is claimed.

Credit

Cinna Davis / 1337isnot1337 owns PR4470's construction. Retain zxv's receiver-sharing, native/controller/compiler work and exact PR4436 navigation repair; Jacob Alan Sussman's mixed-target recursion; David Ganz, Rohan Arun, Marius-Paul Dumitrean, Erwan Secousses, and the full inherited scalar, Clifford, geometry, BIT, precision, and outer lineage. Mark / Matarener's PR4457 is methodological precedent; Mihir Srivastava's PR4504 and Ryanson Jonathan's PR4478 provide complementary coverage. See vendor/NOTICE.md.

The CI adapters retain James Reed's PR3563 finding, the locale package's original checks, the D8 package's authorship and source hashes, and the shared checker contributors. Prepared for Chris / Clear1nsight with substantial OpenAI assistance, Apache-2.0. Separately implemented AI-assisted checking does not mean independent human verification.

Open next questions

The smallest useful follow-up is a source-bound coverage join between PR4504's independently reconstructed chronological obligations and this PR's exact physical recipes. Canonicalise subspaces and orientations; build the required set from events rather than from the list under audit; reject an omitted pair, wrong phase/orientation, and duplicate productive ownership. Retained warp and owned-centre obligations must be classified separately.

Then audit the complete 632-input native invocation and actual p16 copied-query/owner/erasure protocol. These are not closed by the pairwise matrices or a formal controller diagonalisation. TASK.json records the relevance, first step, and blockers. The proposed primary construction, ceiling, native-verification, and compiler-verification assignments are linked above and remain unclaimed.

Workflow feedback (optional)

The original shared plan failure was unrelated to the mathematical audit, but correcting the navigation exposed historical packages that had labelled changing guidance as immutable live scientific inputs. The repair separates historical provenance, frozen replay inputs, and live regression inputs without replacing old hashes or suppressing tests. Future packages should state these roles at creation to keep frontier updates from invalidating unrelated finite research.