[1337isnot1337]

After #4971 replaces three-ADD swaps with field exchanges carrying their frame labels, it leaves open whether chart rescheduling can reduce the paid recursive profile. This package proves that changing at most two intact source-defined constant-frame regions cannot improve the moment for any 0<t<1, with the scalar chronology, literal signed native maps, center interfaces and endpoint charts fixed. No new κ.

The exact graph analysis checks both orientations, each with 768 free regions and 1,474 adjacent pairs. It exhausts 59,231 candidate assignments / 2,090 feasible assignments per orientation. Majorization and three written power inequalities certify every feasible comparison throughout the exponent range; nonadjacent pairs separate. None of the 1,438 identity-edge adjacent pairs admits constant-chart fusion. The source's full {1:2201,2:412,3:510,4:127,5:40,8:8} paid profile, 632 arbitrary inputs and 408 helper returns remain explicit.

The proof, source pins and compressed complete witnesses are in research/swap-contact-fusion-gate/. Both registered Python 3.12.15 modes pass, including -I -O, signed operator/frame checks and rejection controls. The 18-file committed-byte preflight and official export integrity check pass. An optimized standalone replay of the exported bundle also passes with PATH=/nonexistent; all 19 export files stay byte-identical. Hosted CI also passes at exact head 5e1b9843f2b04cee1f4771fce2c62d0ac551d3d6: planner and registered replay jobs succeed.

This does not exclude splitting a source zero-edge region, changing three or more regions, scalar rewrites, frame descents, new centers or changed native operators. Fresh physical phase lowering, full supplier/compiler/BIT/outer admission, Lean and independent human review are absent. The checkout is shallow, so full-history ancestry auditing is not claimed. The selected κ remains 0.00302015.

Scientific base b91956876e5b4122fd89179f1b57557ff8378f15; submission refreshed canonical main bee95ce88f9f19802f0613f92838895cd0e14a1f, with relevant guidance and consumed science inputs unchanged. Scope changes from the rank-shape card and concrete next questions are recorded in TASK.json. Submission overlap refresh covered 302 open PRs, 4,090 discussion records and all 6,636 tracked local note paths; this is bounded coverage, not a novelty guarantee. Only this isolated research package changes.

Credit Dugongue for #4971 and #4935/#4948 source binding; Cao Yifan/intqwq for the round-three source; Cinna Davis, zxv, Jacob Alan Sussman and all retained NOTICE authors; James Reed/jwr404 for #4571 and the independent #4971 review, and #4979 for the operator/frame guard. Prepared by 1337isnot1337 with substantial OpenAI Codex assistance, solo, no subagents. Apache-2.0.