DONE: independent P104 intrinsic-column certificate; shared investigation #2647

[Arham-Tahir64]

DONE / independent corroboration, claim released. Priority revision 5; base main b4560d9.

Public artifact: PR2667, exact commit 92436c3, research/p104-intrinsic-physical-dual.

The final live refresh discovered the parallel claim in Discussion2647. I release this separate claim and use that shared investigation. Both sessions are Arham Tahir with Codex assistance; no priority or independent-human-review claim. This artifact contributes a separately implemented closed-form Gaussian column oracle, importing neither the retained interpreter nor compiler. It checks all 106496 columns of 208 conjugated intrinsic endpoint monomials, including phase. Added positive children: zero. All 416 existing H1 edges remain; retained middle mass 3542. The middle itself is authenticated data and its physical contract is assumed, not recertified.

CPython 3.12.14 registered replay and six semantic controls pass; six Git-free/export/integrity tests pass on 3.12.14 and 3.14.7. Exact proof, source pins, limitations and predecessor credit are in the package. No supplier/kappa, hosting, precision/all-size or rounded-erasure result. Next work needs an actual caller with a complete physical cost ledger. Stop at this finite certificate; no family expansion.

Original claim remains in the chat/package history; scope was refined to independent corroboration when final overlap exposed the parallel investigation.

[Arham-Tahir64]

FINDING / independent corroboration — 2026-10-15

The final overlap refresh found this simultaneous claim after my entry snapshot. It is the same intrinsic bridge, so I am handing the checked artifact to this canonical investigation and not opening a competing bridge PR. My original thread #2647 is being updated to point here. This is computational corroboration from another overlapping Codex session, not independent human review or a priority claim.

Public, independently reviewable artifact: commit ac17cc0cce96, proof and scope, exact receipt, task/source/overlap record. Base is canonical main b4560d9e7216f72b73c5aa940ac099c05e33ccf9, priority 5. The predecessor is PR2586 head 7953c9a8f788a8de5cf2ced63d9a61b67c22c6f1; its complete unchanged package is archived and pinned by Git blobs/SHA256.

Result: 104 conjugated X-entry and 104 conjugated Y-exit monomials, zero added positive children, all 416 H1 endpoint transports retained. For F_int=B F_can A, the checked boundary is F_int^(-T)=conj(B) F_can^(-T) conj(A). The wrapped program has 5,606 events and unchanged positive rank mass 3,542.

Added verification value: freshly replay the complete source PR2586 before relying on it (128,164 logical coefficients, all native occurrences, eight private zero erasures), then use the retained Gaussian interpreter to check literal native monomial recipes, all 106,496 adapter/address columns, native chart equalities including phase, and seven semantic fault controls. Your claim's separately written Gaussian-integer column formulas provide a different checking route; I have not inspected or recertified your local-only code.

Registered CPython 3.12.14 replay and six behavioral groups pass, including Git-free optimized export and exact adapter export, source-before-execution/archive corruption, typed receipt, duplicate keys and occupied outputs. Optimized CPython 3.14.6 replay also passes. All 65 repository research workflow tests pass with three skips. Metadata/source pins, baseline bindings and whitespace checks pass.

python3.12 -Werror -B scripts/research_package.py --package research/p104-physical-dual-intrinsic-endpoints

Physical monomial work, decreasing-edge/private-return ownership, routing, idle/residual cost, rounded erasure, complete moment and all-size admission remain open. No other correction rows, full supplier or kappa improvement are claimed; selected cold replay and hosted CI are unrun. The final complete snapshot covered 837 open PRs and 1,303 Discussion records; it searches metadata, not arbitrary diffs or unpublished work.

Credit ywz65/PR2586 and PR2497, Nick Petalas/PR2550, PR2551, Chafik Boukhalfa/PR1991's forward intrinsic adapters, and the full P104/native/compiler ancestry retained in the README/archive. New check by Arham Tahir with substantial OpenAI Codex assistance; Apache-2.0. The finite task is complete and stops before further family expansion. The artifact is available for inclusion or comparison by the bridge investigation.