[aliromman]

Corroborating replay of concurrent PR5650, not a distinct construction. PR5650 appeared while this package's final checks were running. The next upstream refresh found the overlap. Both independently prepared constructors emit exactly the same complete program, apart from the schema label. Their full native, physical and invoice receipts also agree. This result is excluded from the donor's ten-result quota.

The scope and complete proof, exact cross-PR comparison, concurrent source pins, original full verifier, and complete receipt are fixed at head 853192167b27d44aea0d4bc2b0d52d695fadc7c6. The shared program digest, after removing only the differing schema labels, is feaf13d7ceed28a05e951a70f09355f1c9dca4a27ed5b81b43abb0cbe5a74e63.

The common construction composes zxv's paid A2 exchange (PR5643) with Ali Romman's three helper cleanups (PR5644), both on PR5627. Both application orders reproduce every literal instruction. The full bill retains 743 independent inputs, 12,958 productive targets, 66,492 paid calls and rank surplus760. All 565,504 endpoint entries, 2,916 native maps with exact phases, the complete physical caller, both interval engines, compiler and 58 outer conditions pass. Shared frozen engines are reused; this is not an independent implementation of every engine or independent human scientific acceptance.

Both packages enclose the same raw crossing (0.00310644469581,0.00310644469582). PR5650 claims .00310644468; this separately completed replay tests and supports .00310644469. The extra decimal place is not counted as a mechanism or research improvement. The same five framework assumptions and helper endpoint contract remain, and the selected .00310045 record is unchanged.

python3.12 -B scripts/research_package.py --package research/p104-a2-cleanup-composition
python3.12 -I -O -B research/p104-a2-cleanup-composition/compare.py

Normal and optimized full registered replays and a Git-free full replay passed at bdf59f7af8049ef226e3c68ee730c2ac12d8a421. The scientific constructor, verifier, pricing code, emitted word and receipt are unchanged by this reconciliation. The added comparison passes in normal, optimized and Git-free export modes and rejects a missing-event control. Revised metadata, committed-byte and export checks pass. The hosted planner remains a separate gate; current main has a reported missing-navigation-link regression. Full-history ancestry, formal-kernel proof and human review were not run. The huge uniform word is not materialized.

Coordination and reconciliation. Base: canonical main df0137e, priority7. No new construction scope is reserved, and no other PR is changed or closed.

Credit zxv for PR5643 and concurrent PR5650; Ali Romman for PR5644/5627; Chris R/Clear1nsight, ToikY and the A2 audit lineage; flyingviper, Scratch-net, Pierre-Loic Doulcet, GamingPuzzled, DreamingOfClouds, thu-nmrc, David Arias, Gareth Davidson, Cinna Davis, Jacob Alan Sussman, Cao Yifan and all retained native/geometry/BIT/compiler/analytic notices. This separate construction, replay and comparison used substantial OpenAI Codex assistance for Ali Romman; no subagents. Apache-2.0.