[aliromman]

Compose the released PR5786 asynchronous hubs, thirteen PR5774 retained native copies, Joel Pulikkan's original R22 block, and the newly released PR5796 second block. The complete conditional saving is 0.00312289583377, a gain of 0.00000024539377 over PR5786. The raw crossing is strictly inside (0.00312289583975,0.00312289583976). This strengthens the first published checkpoint of this same PR; it is not an additional counted contribution.

Every source mechanism retains its authorship. Eight exact R22 edits preserve scalar transfer matrices. The second block changes 102 contacts on nine roles while preserving every scalar coefficient and its order. Its integration changes only 16 required native edges and preserves unrelated query paths. A new chart 1092 avoids overwriting an existing basis. All 23 COPY events, including the thirteen retained native copies, keep their logical values. Construction and compatibility proof · literal edit witness · complete result.

The full caller has 13444 productive targets, 66942 paid calls and rank surplus 760. Every native copy, query, adapter, normalizer, scalar operation, owned lifetime, route and idle remains charged. Both rational moment engines, the finite compiler, 54 strict outer rows and four functional conditions pass. With the retained reserve 1e-12 and tested-to-published gap 5e-12, the claim is the maximum on the 1e-14 grid: the next tested point fails the reserve, and a directed derivative bound excludes all later points through the retained BIT cap .0034. This is a fixed-contract optimum. Different compiler parameters, words and caps remain open.

Validation: the primary full replay passed in143.90 seconds. All four registered normal/optimized full and compatibility commands pass in247.49 seconds. The checks cover all795 scalar input columns,632025 coefficients, separate transpose,2926 native maps with absolute phases,27 owned lifetimes,208 main adapters,1124 helper normalizers and the full physical caller. Twelve word/caller/invoice controls, the native phase control, the scalar-transfer control and nine actual frozen-source corruptions reject. Committed bytes/modes and official export pass. The sealed read-only Git-free export passes both scripts under -I -O -W error -B with empty PATH in76.00 and.24 seconds; all31 files retain their bytes and modes. Validation record.

Run python3.12 -B scripts/research_package.py --package research/p104-copy-block-composition.

The published saving leaves5.98e-12 to5.99e-12 below its raw ceiling. This is a small conditional gain; all five framework assumptions remain. No large uniform word or rounded multiplier is materialized. The original fixed R22 chart theorem and the compiler-grid theorem do not prove a global optimum over joint retained-copy, temporal-target or changed-caller constructions. The active forward/nonnested-copy and temporal-cut searches remain separate. Full-history ancestry checking is unavailable in this shallow checkout. Hosted CI and independent scientific acceptance remain pending. No merge, closure or selected-frontier change.

Base and refreshed main: de5ec2e7a6074068452f45b1a191b5c72832df4a, priority8. Canonical investigation. Exact immediate source: PR5786 at 6b936acd68a7e0f9062037900f0fe9bc4b6fdd25; R22 edits: PR5739 at b93d749a0a895d260e87887170af5162bda461a4; second block: PR5796 at 4b30391e633ba0c7c674f03586ae97d443e9bc9c. Credit Joel Pulikkan/GamingPuzzled, the second-block author, PR5786/5774/5751, flyingviper PR5718, Cinna Davis, Jacob Alan Sussman, zxv, Cao Yifan and every retained P104/native/BIT predecessor. New integration, compatibility and fixed-grid proof: Ali Romman with substantial OpenAI Codex assistance; no subagents. Apache-2.0.