Certify the staged544 single-ADD retiming obstruction and a paid circuit control

[Clear1nsight]

Change and mathematical scope

Adds only research/rank-profile-retiming-obstruction/: an exact single-ADD retiming census on both literal PR3734 staged544 providers, a scoped all-exponent non-improvement proof and a credited paid-circuit control. Original fork PR: Clear1nsight#9.

Every transformation in the declared class has nonnegative new-minus-old raw moment for 0<a<1. The relaxation covers 1776295 insertion-cut/frame-dimension candidates per provider and 19 signed histogram types with exact certificates. It excludes #3863's proposed 2+2 -> 1+3 exchange through this class only. One existing ADD may move within A, B or restore, after the A prelude and without crossing native B warps. Crossed ADDs satisfy the sufficient commuting condition; nonzero SCALEs are conjugated exactly. Any binary common-frame subspace is allowed; all other gates/frames/entrances/cuts/exits are fixed. There is no radius bound, and original-position frame-only changes are covered. Relaxed virtual-climb boundary charges vanish for genuinely legal witnesses.

The constructive control rolls back and recovers an already-retained Jeremy Kothe #2957 swap: 2+3 -> 1+4 with no extra scalar gate, helper, private cleanup buffer or helper-width change. Both providers recover the literal pinned word and pass the retained exact free-operator interpreter in both orientations with 632 live inputs and 91 arbitrary dirty buffers. It is a local counterfactual rollback, not a reconstructed historical parent or new present improvement. #3863's credit theorem gives 1491695107/1513529160000 per call on 17/10000<=a<=18/10000; at 3240 repetitions the reserve exceeds three rank-one children by 2437538889/12612743000, conditional on unchanged normalisers and the complete common invoice. Those extra operations are not newly implemented.

This is a scoped negative result and paid control, not a global optimum, supplier admission, outer release, new kappa or selected-frontier change.

Task and overlap

Assigned rank-profile circuit-transformation activity; refined to single-ADD retiming with all other interfaces fixed. See TASK.json. Canonical base: 20ecf28; priority revision 6. Scientific head: 359fe0f. The comparison is one commit and seventeen files, all in this new package. Current AGENTS.md, autoresearch.md and CONTRIBUTING.md were read for submission.

Pinned dependencies: dataisfire/#3734 fa04103; Dugongue/#3863 635fe84. #2957, #2979, staged-entry/retiming and ranked-pair-verification retain their roles. Added value is the full declared-class census and exact certificates against the literal frozen providers, plus the source-recovery control. The original investigation checked live relevant PRs; Discussion HTML coverage was partial and find_overlap.py was not run. No exhaustive novelty claim is made. Frozen predecessor evidence and temporary fork-only export workflows are not changed/included.

Validation

python3 -B research/rank-profile-retiming-obstruction/verify.py --summary
python3 -O -B research/rank-profile-retiming-obstruction/verify.py --summary
python3 -B scripts/research_package.py --package research/rank-profile-retiming-obstruction

The first two commands passed in the original local CPython 3.13.5/Linux x86_64 standard-library environment in 26.24s and 33.34s. Each authenticates retained source bytes, independently reconstructs chronological invoices, regenerates both full censuses, checks all sign certificates, recovers the literal control word and runs eight rejection/sensitivity controls. Wrong-scalar controls refresh cleanup before rejection. Exact proof, census, source pins and reproduction environment are retained.

The Python 3.12 registry hook is supplied but was not run locally in that record; metadata/history validation and hosted CI are separate. This browser publication performs no new scientific execution. Fresh native Clifford, physical allocation, stock/precision, full supplier/outer release, all-size/compiler, analytic and BIT checks were not run. Neither independent human review nor formal-kernel acceptance is claimed.

Credit

Credit dataisfire/#3734, Dugongue/#3863, Jeremy Kothe/#2957, #2979 and the retained staged-entry/retiming chain. Sagar Thakrar's auditor and roger-sato's adapter, including Arham Tahir's frame-contract attribution, are reused unchanged and are not described as newly independent interpreters. New work by Chris R / Clear1nsight with substantial OpenAI assistance. Apache-2.0 and all retained source/interpreter notices apply.

Open next questions

See package README.md and TASK.json. Test a neutral first retiming followed by a coordinated second gate/frame edit, or construct a literal boundary-changing implementation of #3863's rank-two exchange. First supply before/after words and a complete arbitrary-dirty restoration and physical cost ledger. Coupled changes, multiple frames, warp crossings and changed interfaces remain open. The frozen-baseline certificate cannot be iterated into a universal obstruction; unknown whole-circuit costs remain blockers.

Workflow feedback

Upstream creation through the integration previously returned HTTP 403. This separate browser submission transfers the original focused contribution while retaining fork #9 as its review record.