Map every source row to its whole complement invocation and child input slot without expanding replicas

research/source-complement-port-index/NOTICE.md

@@ -0,0 +1,22 @@
+# Notice and retained credit
+
+New true-rank local count reduction, source-bucket indexing, whole-complement
+port map, proof and checker: prepared for Jerry Chen / @CubeJerry with
+substantial OpenAI assistance, 2026. Apache-2.0.
+
+The exact source-row index at CubeJerry commit
+c5bcb6dc792f5fe4d8abe89a2a2854f148b09f2c is retained byte-for-byte under
+inputs/source_index/. Its unchanged Pascal codec originates at CubeJerry
+commit 91246e5131152ccb1f64a969c36706360861f821. Their headers and independent
+review lineage remain credited.
+
+inputs/finite_dag_audit.py and the reference notes are exact canonical e545
+bytes with original notices. Jacob Alan Sussman supplies Theorem 1G and its
+Pascal flow; zxv supplies the native finite compiler and literal DAG source;
+Mihir Srivastava supplies chronological-batch-width (#4507); JoyboyBrian
+supplies symbolic-integer-batch-width (#5071). Their source formulas,
+batch-factor minimality and operator hypotheses are inherited results.
+
+The checks are AI-assisted author evidence, with any separate peer review
+identified separately. No human scientific review or formal-kernel
+acceptance is claimed.

research/source-complement-port-index/PR-TITLE.txt

@@ -0,0 +1 @@
+Map every source row to its whole complement invocation and child input slot without expanding replicas

research/source-complement-port-index/PR.md

@@ -0,0 +1,94 @@
+## Change and mathematical scope
+
+The published [source-row index](https://github.com/CubeJerry/integer-mult-bounds/tree/c5bcb6dc792f5fe4d8abe89a2a2854f148b09f2c/research/lockstep-source-row-index)
+left source-rank grouping and complement batches open. This supplies an
+exact inverse map between its replicated row integers and every whole
+complement invocation/input slot in FINITE-PRECOMPILATION section 4.
+Sources of rank N receive an explicit no-complement label; no B_0 is invoked.
+
+For true job rank s, freshness simplifies to A*s>N*r_last and unit fit to
+A*s<=N*k_i. Shared product tables then supply each class's rank-s source
+count in d lookups. Each bucket retains occurrence order and supports exact
+rank/unrank. Given the supplied complete smaller-program width B_(N-s),
+quotient/remainder of b*S_s+j selects the invocation and its input slot.
+The inverse recovers the original replica-major source row.
+
+**No new kappa, width improvement or full assumption discharge.** Selected
+conditional kappa remains 0.00363005817.
+
+## Why this is new
+
+#4507 and #5071 already prove the source census, whole-batch divisibility and
+least gcd/LCM replication factors. Those results are credited. This adds the
+individual source-occurrence/child-port identification without expanding
+all replicas, unit ordinals or class paths. Equal true ranks and coincident
+products share counts while retaining separate source fields. The complete
+deterministic metadata cost includes candidate ranks with zero sources and
+a paid serialized-table access model.
+
+## Verification
+
+```sh
+python3 -I -W error -B research/source-complement-port-index/verify.py
+python3 -I -W error -O -B research/source-complement-port-index/verify.py
+```
+
+Normal and optimized receipts agree: 26 literal flows, 755 sources, 1,334
+replicated rows, four fitting product-collision classes and 18 rejecting
+controls. A source-pinned literal DAG oracle checks every bucket/port inverse.
+A height-24 case indexes 216,192,013,559,808 sources per replica with 601
+states and 266 candidate ranks. An independent target-minus-incoming census
+checks all 206 nonempty rank counts; six large replicated positions round-trip.
+
+A concrete rank-5 width-three complement uses old rows (2,3,6), then
+(7,10,11), crossing replica boundaries. Full-rank rows (1,5,9) skip the
+complement. Mutations reject incorrect freshness equality, overwritten
+product coincidences, merged paid sites and merged replicas.
+
+Canonical base: e54523517d0f03e58352e647c403a5af3e525a57.
+Source-index dependency: c5bcb6dc792f5fe4d8abe89a2a2854f148b09f2c.
+Occurrence-codec origin: 91246e5131152ccb1f64a969c36706360861f821.
+Exact input blobs and byte hashes are recorded in SOURCES.json.
+
+## Limitations and next obligations
+
+This is a logical grouping/dispatch interface. Payload permutation, adapters,
+copy/restoration/erasure, actual smaller programs and the selected native
+manifest remain supplied obligations. Complement divisibility does not
+replace separate chronological exit-cut constraints. No owned bank becomes
+an independent productive source. The complete multiplier, five framework
+assumptions and all numerical/precision/tape costs remain conditional.
+
+Next: bind one real smaller-program library and whole-field adapter to these
+ports, then price actual grouping and return. Separate exit-port indexing
+must preserve its chronological cut boundaries.
+
+## Credit
+
+Prepared for Jerry Chen / **@CubeJerry** with substantial OpenAI assistance.
+Credit the two published codecs and their separate AI reviews; Jacob Alan
+Sussman's integer Pascal flow; zxv's native compiler/literal oracle;
+Mihir Srivastava's #4507; JoyboyBrian's #5071; and the cited adaptive-width
+predecessors. Original source headers are unchanged. Apache-2.0.
+
+## Independent acceptance checks
+
+The orchestrator read the full proof and implementation, checked the exact
+source substitutions, candidate-rank bound, interval inverses and the complete
+metadata cost distinction. The separate [peer review](independent/REVIEW.md)
+constructs expected sources by literal role-word parent edges without using
+the author census, freshness predicate or parent resolver. It passes 22 flows,
+328 sources, 3,900 replicated rows, 2,707 whole invocation rectangles, 51
+NoComplement rows and 11 rejection controls. Both cross-replica batches are
+recovered independently. A minor transient old/new bucket wording correction
+was incorporated; the executable files were unchanged.
+
+```sh
+python3 -I -W error -B research/source-complement-port-index/independent/check.py
+```
+
+Expected: `PASS_SOURCE_GROUPS_PEER_REVIEW cases=22 sources=328 rows=3900 controls=11`.
+The exact reviewed hashes and test domain are in `independent/RESULT.json`.
+Author normal and optimized replays pass identically from this final package.
+These are independent AI checks, not human or formal-kernel acceptance.
+The complete proposed PR description is committed in `PR.md`.