[unpaidmoon]

Change and mathematical scope

Independent scalar audit of the four delayed helper-input lifts in merged PR #6349, head af7cb6c67f8fc7d82a2e4aee2dd2f07481bf6690. The source word is PR6314. This is a scoped audit, not a new construction and not a kappa improvement.

A separate replay of both pinned words accepts every scalar endpoint column. The four input directions, their inverse output additions, and both dirty-carrier sandwiches match the published schedule. Helper ranks move 9→8, 9→8, 9→8 and 9→6, so the local invoice changes by paid H1+H2+H3+2H4-4H5 and productive H6+3H8-4H9.

Native phases, the physical caller, both price engines, the compiler threshold and the outer predicates were not run. The five framework assumptions remain. Selected kappa stays 781451/250000000.

Task and overlap

Base commit: 7c49fe32c55de02c2c873655c804b16e4c3fef0a. Priority revision 8. Audited head: af7cb6c67f8fc7d82a2e4aee2dd2f07481bf6690.

Task card: p104-side-correction. Scope status: changed. The card's first experiment starts from selected #5599. This package instead answers discussion #1166 comment 19341394, the review of these four lifts on the PR6314 word.

Nearest work:

• Ali Romman, PR Lift four P104 helper inputs: conditional κ 0.003129111 #6349: the construction under review. Archival merge is not scientific acceptance. This package reimplements the scalar endpoint, frame inclusion and histogram comparison and does not import that checker.
• PR Compose native input lifts with the query tree at conditional κ 0.00314980 #6350 and PR Contract nine native helper outputs at conditional κ 0.00315057 #6353: later compositions and output contractions. Their audits do not check this head's carrier sandwiches.
• DreamingOfClouds, PR No new κ: shared partial sums and row sharing cut PR6167's word from 22,821 to 17,445 events #6314: the pinned source word.

Overlap refresh at 2026-11-02T13:30:50Z covered 321 open PR records and 6184 discussion records. No open PR performs this audit.

Validation

Python 3.12, standard library. From the repository root:

python3 -W error -B research/p104-input-lift-direction-audit/verify.py
python3 -O -W error -B research/p104-input-lift-direction-audit/verify.py
python3 scripts/research_package.py --package research/p104-input-lift-direction-audit --check-only
python3 scripts/research_package.py --package research/p104-input-lift-direction-audit

Both registered commands passed locally and print PASS_P104_INPUT_LIFT_DIRECTION_AUDIT. Each scientific command takes a few seconds. Four mutations fail closed: a missing carrier subtraction, a missing output correction, a reversed 649 injection, and a frame-1093 basis that is not the span intersection of frames 118 and 733.

Not run: the predecessor native checker, caller replay, rational price engines, and the registered PR6349 verifier (timeout_seconds 600).

Credit

Audit prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0.

Construction and author checks: Ali Romman, with substantial OpenAI Codex assistance. Source word: DreamingOfClouds, with Claude Opus 5.5 assistance, in PR #6314, retaining PR #6167.

Open next questions

The signed Pauli image of one changed helper normalizer is still unchecked. That is the smallest native gate for this word; native_all_columns and the predecessor checker are the blocker. PR #6353 needs its own initial-frame reading before these four ranks are reused. Details are in research/p104-input-lift-direction-audit/TASK.json.

Workflow feedback (optional)

The handoff names one registered verifier whose native and caller sections are a different obligation from the scalar schedule. Splitting those gates in the review request would make the 60-minute scope obvious.