[bitplane]

Seven complete six-register preparation blocks in the frozen PR936 word admit shorter Boolean-equivalent programs. Replacing them and rebuilding the literal reverse cleanup reduces the complete local circuit 23,711 → 23,673 ADDs, saving 38 more ADDs. Against selected PR589, the cumulative saving is 632, including the 254 from Gareth Davidson's PR904 and 340 from Ross Richardson's PR936. Finite construction; no new kappa.

The forward/reverse positive bounds improve 41,311 / 843,188 → 41,287 / 833,084. Local payload remains 92 bits under 104; the inherited nine-stage expression needs 174 bits under its 192-bit charge. Original H1–H5, every register's ordered MOVE path, all COPY/ERASE records, 439 dirty helpers, nine rank-seven copies and paid rank mass 5,129 are preserved. Complete 5/9/11-stage savings beyond PR936 are 190 / 342 / 418 ADDs, with unchanged bridges and copies.

The bounded scan covers all 60 six-wire A blocks, 720 pivot orders, two donor priorities and row-only/mixed row-column elimination: 172,800 nominations, 15,469 distinct shorter programs and 440 passing both positive guards. Seven block winners save 19 A ADDs plus 19 C ADDs. Five change the positive integer map while preserving the Boolean map; two preserve both. This whole-block class differs from PR904/936's small windows and PR925's fixed positive-transfer minimum on frame 267, which remains untouched. No unrestricted optimum, fixed point, necessity of six wires or exclusion of another five-wire improvement is claimed. No D-compensation or other B-block savings are composed here.

Reproduce with Python 3.12+, C++17 and Boost headers:

BOOST_INCLUDE=/path/to/boost-headers python3 -I -B research/r439-six-wire-boolean/verify.py
python3 -I -B research/r439-six-wire-boolean/test_package.py

Expected: PASS_R439_SIX_WIRE_BOOLEAN: 7 blocks; 38 further ADDs; 632 versus PR589; H1-H5; 5/9/11 stages; 12 controls; no new kappa.

Full receipts match on CPython 3.14.4 and a minimal read-only offline copy under isolated optimized CPython 3.12.3. The independent local checker covers 448 Boolean inputs and 84 positive-column pairs. A separate set-valued interpreter replays all 661 dirty-input columns in both directions on three complete words; unchanged source geometry/interface checks and fresh native 5/9/11-stage checks pass on all three (883 / 1,327 / 1,549 columns each). All 144 bridge controls, 12 local controls and seven artifact tests pass. Exact positive bounds and ADD/payload invoices agree across implementations. Selected-record checks pass before and after rebase; all 18 consumed source/predecessor artifacts match their pinned Git objects. Staged whitespace checks pass.

OS seed 19, bit-supplier track. Initial base 47481c8; publication base 0076643 after the maintainer batch932 integration. All thirteen consumed upstream files are byte-identical across that update. PR936 is pinned at 0679fb6, with its original candidate/certificate/attribution retained. Final overlap refresh through PR951 found no duplicate six-wire result and no changed PR936 head. No pinned coordination issue existed; the claim is local. Only research/r439-six-wire-boolean/ is added.

The selected PR890 conditional κ remains 0.001661957774647029. The paid profile and exponent are unchanged. Adoption still requires binding this candidate into the full hosted supplier emitter, rebuilding its invoice and outer assembly, and retaining the compiler, physical label/mutable-source, hosting, rounding, precision/tape and analytic assumptions. No full repository suite, selected complex cold replay, hosted CI pass or formal multiplication proof is claimed.

Credit Ross Richardson PR936/925, Gareth Davidson PR904, Jeffrey Cox PR754, Sagar Thakrar's native harness, DreamingOfClouds PR589, eumemic PR555/543 and the retained Dylan Crow/DaysSky/Zhihao Chen source lineage. Prepared for Gareth Davidson (@bitplane), with substantial OpenAI Codex assistance; one AI agent, no independent human review. Apache-2.0 and predecessor notices retained.

Published head: 11d2470. Mathematical implementation: 2b1f30a; the final commit only records the completed local claim and PR link. The OPEN state, exact remote head, 20-file scope and clean worktree were verified. Hosted CI is pending: Graded box convolution=queued, Verify early reset and kernel composition=queued, Verify terminal elimination on the paired-cube bit word=queued, Verify research artifacts=queued, Verify recycled bit registers on the paired-cube bit word=queued. No hosted CI pass is claimed. Research-artifact workflow.