Bind pooled R437 moves to retained IR emission and expose replica catalogue boundary

[Clear1nsight]

PR #3959's first cross-replica move cannot enter the retained physical catalogue: stage 6 / replica 1 / helper 421 changes family 1097 / offset 0 to family 209 / offset 16, violating its exact replica symmetry. This package executes that actual guard, supplies an absolute-owner IR adapter for all 63 pinned moves, and identifies the required replica-indexed native interface. Partial result: finite emitted-IR binding; pooled physical CALL/RETURN remains open.

The unchanged retained R437 stream-emission block and the separately retained audit encoder agree on all 7,814,016 local events. Exactly 2,265 events change. The new stream digest is 68d691118d9ac43f79a35cd2d6662a8257e68eddfc6a0fd79524ca15145c775e. All 125,856 hosted roles and 63,936 replica-indexed idle records are checked against literal chronology, absolute families, extents, disjoint cells and compact prefixes. The credited #3959 profile delta is reproduced; paid requests 903,456, rank mass 1,808,352 and maximum child 42 are retained.

Removing the old guard alone is insufficient: its canonical-request count multiplies every rank count by 32 and its semantic key omits caller replica. The pooled profile contains 400,877 rank-2 and 1 rank-14 requests, so that factorization cannot represent it. The first-move witness is a compiler-interface counterexample, not a counterexample to pooling. Six wrong-owner/replica/overlap/extent/pair/duplicate mutations reject.

The proof pairs reversible compiler refinement with network-capacity allocation and compositional transport on products of register sets. The transport lemma covers arbitrary coordinate values and dirty-helper states, conditional on the inherited local word and valid charts. It does not price address routing as free or assert a native execution. Source identities, pooled owner charts, preparation, physical returns, consumers, parking, scratch, visited cells, precision and full supplier admission remain explicit unknowns. No new κ or selected-frontier change is claimed. #3967 and #3980 are credited, explicitly unpooled predecessors; the packing and conditional endpoint remain Rohan Arun's #3959 result.

Reproduce: python3 -B scripts/research_package.py --package research/r437-replica-indexed-emission. Registered replay passed; the source-bound checker uses only the standard library. Exact source files are retained and checked against their original Git objects, including ai-nanachi's #3647 and Rohan Arun's #3959. The package contains source pins, proof, result, meaningful controls and open obligations. Neither the two retained encoders nor this same-session work is presented as independent human or formal-kernel acceptance.

Main 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority revision 6, refreshed unchanged before publication. The parent checked 679 live open PR bodies and active/handoff discussions, then #3980/#3982/#3983 deltas and a complete 691-open-PR checkpoint at 08:13 UTC; the local overlap helper examined 2,492 records with no usable remote cache. Wider Discussion, closed and private coverage is partial. Explicit donor assignment refines supplier-boundary-audit; no new packing census was performed.

Credit Rohan Arun (#3959/#3958/#3939), ai-nanachi (#3647), Ali Romman (#1043), Sagar Thakrar (#3700/#2423/#1023), mightyOil (#1281), DaysSky, chafreaky, DreamingOfClouds, eumemic and the retained E8/hosting/emitter lineage. Physical catalogue lineage includes zxv, David Arias, Jose Manuel/jmbarrancoml, ywz65, Nick Petalas, alephnan, Adam Gall and Ross Richardson. Full notices retained. New work by Chris R / Clear1nsight with substantial OpenAI Codex assistance, Apache-2.0.