[aliromman]

Use the earlier helper value to remove one internal contact. This raises the complete conditional κ from 0.00310644469 to 0.00310676333, a gain of 0.00000031864 over PR #5651's corroborating replay of PR #5650. This is a small construction gain.

The starting word is the released PR #5643 / #5644 composition from PR #5650. Its exact input is retained and checked. The new four-gate schedule replaces h-=4s; c+=2s; c-=x/2; x+=h by c+=h/2; h-=4s; x+=h; c-=x/2, at the specified legal native frames. The local paid change is −2 H1 + H2, with no target loss. The complete caller removes 18 paid requests and retains 12,958 productive fields and rank surplus 760.

Quantity	PR #5651	This contribution
Complete paid requests	66,492	66,474
Productive fields	12,958	12,958
Conditional κ	0.00310644469	0.00310676333

The unique raw crossing is (0.00310676334487,0.00310676334488). The tested saving is 0.003106763335, with reserve 10^-12. Every local/helper/native map, adapter, thin query, idle, owned operation, restoration and route remains charged. The constructor changes the scalar graph; it does not merely choose more decimal places below an old invoice ceiling.

The package includes the exact witness, complete emitted word, source pins, proof, complete receipt, explicit normal/optimized replay hook, finite pattern census and corruption controls. At scientific commit 1f5a93245540e79be6ff4e7d573bbaf823a9dc23, the complete registered normal and explicit optimized replays pass. They check all 565,504 endpoint coefficients, 2,917 native maps and exact phases, the physical caller, two interval engines, compiler costs and 58 outer predicates. The official export also passes from a read-only directory with no Git and PATH=/nonexistent; all 23 exported files keep their bytes and modes. Six actual one-byte source corruptions reject before source import and are restored exactly.

Publication head is 3b0b4f37d5005f381ed9cfcccebfbe1ab35c71aa. It changes only validation/proof prose and adds the corruption receipt after the full replay. All scientific code, inputs, emitted word and complete result remain byte-identical. Metadata, source pins, committed bytes and official export integrity pass again at this head. The README and proof were rendered through GitHub Markdown and their table/code text checked. Full-history ancestry is unavailable in this shallow checkout. Hosted CI and independent scientific review remain separate; current main has a reported missing-navigation-link failure in the shared planner.

Proof · Exact witness · Complete receipt · Validation and limits · Source pins.

python3.12 -B scripts/research_package.py --package research/p104-contact-rewrite

The same five selected framework assumptions and helper endpoint contract remain. The huge uniform word and complete rounded multiplier are not materialized. No global maximum, large breakthrough, proof-kernel theorem or independent human acceptance is claimed. The finite pattern census finds this one pattern before the change and none afterward; other dependency orders and longer identities remain open. Selected result 0.00310045 is unchanged, and no PR is merged or closed.

Credit zxv for PR #5643 and the concurrent PR #5650 composition, Ali Romman for PR #5644 and the PR #5651 corroboration, Chris R/Clear1nsight and ToikY for A2, and Cinna Davis for compensated triangles and scalar compression. All P104, relay, preparation, geometry, native, BIT, compiler and analytic source notices are retained. The new four-gate placement and author checks were prepared by Ali Romman with substantial OpenAI Codex assistance; no subagents. Apache-2.0.

Canonical coordination. Base canonical main df0137efd35e11b4445734bf612bdd4c7ebc330c, priority 7. The stronger headline in PR #5440 has an explicitly unresolved native chain-to-event attachment; it is a separate conditional scope, not a completed replacement used by this package.