Scoped negative result: partial K/M face bypass requires four carriers

[Mister-Meeseeks]

Scoped negative result / research note; no new κ. A mixed K/M face bypass appears to remove two copy helpers by letting targets read raw leavers. Keeping the ordinary consumer's later delivery forces four carriers of the remaining packet. With the other register roles fixed, two extra copies cancel the helper saving. A fully reconstructed attaining control has 492 helpers and a strictly worse paid moment than its raw K492 baseline for every exponent 0 < s < 1.

The new evidence is confined to research/partial-k-face-bypass/:

• Exact geometric witnesses cover all 16 mixed-pair/sign cases, with a short monotone-frame argument deriving packet-last while allowing arbitrary own-read order.
• Complete memoized searches for seed-selected pair (0,2), both signs, exhaust two carriers (63,383 states) and three (156,085 states); four carriers have a concrete schedule. Omitting the ordinary-consumer guard produces a recorded false positive.
• The paid negative-sign control passes the unchanged native auditor on all 716 columns in both orientations, including dirty compensation, inverse cleanup, source accesses and frame accounting. Its exact profile difference is +10 H1 − 6 H2 + 2 H3 + 2 H4 − 2 H6; integer majorization proves the strict moment loss.

Research note and proof, exact evidence, source pins.

python3 -B research/partial-k-face-bypass/verify.py

Expected: PASS_PARTIAL_K_BYPASS_FLOOR_AND_UNFAVORABLE_PAID_CONTROL.

Local verification passes on Python 3.12.8 and 3.14.6, including isolated optimized execution and relocation to a minimal checkout. Four circuit mutations, changed inherited source and corrupted expected evidence are rejected. Separate explicit XOR-span enumeration agrees on all 64 target spaces. make selected-record-check, package hashes, documentation links and staged whitespace pass. No full repository suite, selected cold replay, Linux CI completion or Lean proof is claimed.

Base: 66849e1475c930fd7732b1a0067ffee7e1e95f60. OS assignment seed 24, complex supplier. No pinned coordination issue existed; the local claim records the completed result and scope. The full queue was refreshed through #680 (500 open PRs), with an immediate pre-submission check through #682. This differs from #646's full signed triangles and #651's own-first face/slot sharing by splitting faces and allowing own-read movement. #679's selector-volume batching and #675's owner-boundary audit are distinct. Concurrent #681 resynthesizes a coordinate-total factory window with unchanged frames; this note changes and prices the later face-delivery schedule, so its evidence is also distinct. All existing files, upstream/, and the selected frontier are unchanged.

The carrier obstruction fixes the decomposition, pure packet carriers, other register roles and ordinary delivery. It excludes neither encoded/borrowed/recycled carriers nor other factories, one-face variants, or better four-carrier schedules. The moment comparison concerns the supplied raw control; it does not claim a bound for the optimized selected providers. Global owner admission, phase lowering, optimized kernel/cold-end transport and outer assembly are not supplied. All inherited all-size and analytic assumptions remain.

Credits: icekylinx's #531 factory and ancestry; Jeremy Kothe/AtomScout's #622 leaver construction; Chafik Boukhalfa's #626, eumemic's #543, DreamingOfClouds and the native-audit lineage; Jeremy Diviney's #646, Sreekanth Nair/higherbeing's #651, and Dominik Scholz's #571 packet-plane work. Prepared by Douglas R. Colkitt with substantial OpenAI Codex assistance, under Apache-2.0.

[Mister-Meeseeks]

Integrated at ee055dc44815497927ef618fc985b93c2dcbda2a in the 20-PR pilot, retaining your original commit, source notices and contributor credit.

Fixed mixed K/M bypass: four carriers under stated monotone pure-carrier constraints. Exact geometry and 716-column native control replayed; strict concavity establishes the supplied profile loss, not every possible four-carrier schedule.

The scoped package check passed at cbf30fdb56126d8cb7594f32e4d9a74ea7cd80a7. This is research integration; the selected κ and inherited theorem assumptions are unchanged. Review, evidence and attribution. Thank you for the contribution.