[JoyboyBrian]

Summary

PR4174's exact production fixture cannot supply the requested split-axis two-target color for the PR4990 seven-descriptor frontend.

PR4174 production Layout (65536 axes, alternating widths 24/23)
  all 16 levels × both partitions × exchange cuts
    26 eligible split-axis/batch candidates
      exact image-distance unions: no positive multiple of K=196608

The proof derives the complete candidate set from the actual allocator and allows every bit of each relevant axis, a superset of every synchronized Beneš layer. Therefore no same-color pair containing a split axis exists in this layout, even before full-guard admission. This stops the frontend at its real-caller gate; it makes no claim about other layouts, pairs without split axes or universal CRT feasibility.

Evidence

• Before: The proposed first fixture assumed an actual two-target full-guard color containing a split axis in PR4174's production layout; PR4990 begins with supplied unary geometry.
  After: The source-bound verifier authenticates and extracts the owning production setup, proves candidate coverage, compares independent interval exchanges with Plan.map_position, and checks 57,875 literal bit pairs. All 26 candidates have zero same-color pairs. Inclusive-endpoint and wrong-period positive controls remain active under -O.
• Python 3.12.14: registered normal/optimized replay, metadata/source pins, committed-byte/source-history preflight and whitespace checks passed. The full production census and payload, scanner/repair, native, Lean and Docker replays were not run.
• Credit Hiroki KUMAZAKI (@kumagi) for PR4174/PR4990 and Cinna Davis for PR4100, with the retained ToikY, unpaidmoon, Ryanson Jonathan, Douglas Colkitt and John-Michael Smith lineage. New work by JoyboyBrian with substantial OpenAI Codex assistance. Apache-2.0. Canonical claim: DONE: paid two-target unary CRT stencil and repair join #4985 (comment).

Merge Danger

Door: Two-way.

Blast Radius: Isolated.

Only a new research package is added; predecessors and the selected frontier are unchanged. This finite source-bound obstruction does not implement or price binary descriptor generation, transport, return or erasure. A different actual caller witness is required before restarting that implementation.

Task, overlap and open questions

Refines uniform-transfer-audit / b2_crt_actual_geometry_descriptor on scientific base 4c0a359d88f22f05189d51f48387750b801a45d4, priority revision 7. The originally requested descriptor construction stops because its literal production caller does not exist; the new result is the exact coverage proof, not another synthetic geometry audit. The package README records the next question: find a different admitted caller before implementing its descriptor frontend.

Pre-submission checkpoint: canonical main ff3cc58e3 was fetched on 2026-10-23; governing guidance and selected record are unchanged. Fresh advisory overlap covered 371 open PRs and 4587 discussion records with complete reported pagination. No competing implementation of this precise scoped result was found; arbitrary diffs and unpublished work remain outside that coverage. Official package export/check and an optimized Git-free replay pass. A separate Astra source/proof review confirmed the exact submitted head; this does not constitute human or proof-kernel acceptance.