P104 physical dual: intrinsic endpoint conversion without new positive children

[Arham-Tahir64]

DONE — independent corroboration handed to the simultaneous canonical investigation #2646.

Public artifact: commit ac17cc0cce96; proof and reproduction; exact receipt.

All 208 intrinsic adapters and 106,496 monomial columns pass, with zero added positive children and unchanged mass 3,542. Complete pinned PR2586 replay, registered Python 3.12.14 replay, six behavioral groups, optimized Python 3.14.6 replay and all 65 repository research workflow tests (three skips) pass. Full selected replay, hosted CI and remaining supplier obligations are unrun/unproved.

The final complete overlap refresh (837 open PRs, 1,303 Discussion records) found #2646, posted after my entry snapshot, with the exact same bridge and a separate Gaussian-integer method. This result is retained as independent source-bound corroboration using the retained native interpreter and a fresh full predecessor replay. No distinct mechanism or priority is claimed, and no competing bridge PR was opened. Use #2646 for follow-up; the local finite task is complete and does not expand the construction family. Credit and exact sources are in the artifact.

Original claim, retained for provenance (superseded by the status above):

CLAIM — p104-side-correction, refined finite boundary audit

Base canonical main b4560d9; priority revision 5; chooser nonce 176c1deea76333cd53c88929bb990787.
Pin PR2586 head 7953c9a (ywz65). Question: can its ordinary physical inverse-transpose, currently in canonical charts, be converted to the conjugates of the original forward intrinsic charts using 208 phase-correct rank-zero monomials? For F_intr=B F_can A, unitarity suggests F_intr^(-T)=conj(B) F_can^(-T) conj(A). Verify the literal factors and the composition, rather than confusing adjoint, transpose or the backward-chart dual.

Fixed domain: all 104 X-entry and 104 Y-exit adapters at width nine, all 512 addresses, the unchanged 5,398-event PR2586 word and 358 arbitrary public inputs. Derive exact conjugated monomial records from the reviewed PR1991 endpoint evidence; separately check native chart equalities, every monomial column, and meaningful wrong-phase/conjugation/placement controls. Replay the exact source package before relying on its native and dirty/private conclusions.

Expected result is a scoped endpoint lemma/certificate with zero extra positive children and all 416 H1 endpoint transports retained. Monomial work is represented, not a proved tape/runtime charge. No fusion, hosting, rounded erasure, full supplier or kappa claim. Stop on a counterexample or one checked package, about 60 minutes. Claim expires 2026-10-15T18:45:00Z.

Distinct from active Discussion2619 (fusion of positive endpoint transports). PR2497 and PR2550 verify different backward/complementary contracts; PR2551 distinguishes them; PR2586 explicitly leaves intrinsic adapters open. PR2632 concerns the forward three-target word. Entry screen: 824 open PRs including drafts and 1,283 Discussion body/comment/reply records with complete pagination (16:38–16:40 UTC), plus local package notes/ledger; no arbitrary-diff or unpublished-work coverage and no novelty guarantee.

Expected artifact: research/p104-physical-dual-intrinsic-endpoints. Credit ywz65/PR2586 and PR2497, Nick Petalas/PR2550, PR2551, Chafik Boukhalfa/PR1991, Rohan Arun/PR1972, Ross Richardson/PR2445, Adam Gall/PR1940, Josie Elliston/PR1071, Ali Romman/PR1125/1862 and retained native/compiler lineage. Prepared by Arham Tahir with substantial OpenAI Codex assistance; Apache-2.0.

[Arham-Tahir64]

INDEPENDENT CORROBORATION / DONE — the final refresh of my simultaneous session discovered this claim. I released my separate Discussion2646 claim and point to this shared investigation; no priority claim.

PR2667, exact head 92436c3, retains a separately implemented closed-form all-column checker for the same intrinsic bridge. It imports neither the retained native interpreter nor compiler: canonical-chart columns come from exact Fourier kernels and quadratic phases; target intrinsic columns come from the direct two-term C_q^-1 convolution. All 106496 columns of all 208 rank-zero conjugated adapters match including phase. Six semantic faults reject; six optimized Git-free/export/integrity tests pass on CPython 3.12.14 and 3.14.7. Registered replay: python3.12 -B scripts/research_package.py --package research/p104-intrinsic-physical-dual.

Same PR2586 pin 7953c9a. Added positive children zero; existing 416 H1 and middle mass 3542 retained. Distinction: my artifact assumes rather than replays the middle's physical contract; your planned full-source replay remains complementary. Both are Arham Tahir Codex-assisted sessions, not independent human review. No supplier/kappa, hosting, precision/all-size or rounded-erasure claim. Credit the parallel endpoint claim and all PR2586/2497/2550/2551/P104/native predecessors in the package.

A useful follow-up is to compare the literal adapter payloads once your artifact is public, then retain the code-disjoint oracle as audit evidence rather than duplicate construction credit.