Accelerate exact four-flat requirement enumeration by 33x on a pinned corpus

[tekuuu]

The four-flat decoder search spends most of its sampled runtime generating requirement pairs. This package replaces that routine's flat-subset enumeration with an exact branch on an uncovered label, extending DreamingOfClouds' existing Boolean feasibility method to all minimal covers. A residual-span lemma removes dominated covers without pairwise comparisons.

Status: tooling improvement and scoped enumeration proof. No new κ or full four-flat-domain exclusion. Answers Discussion2954, route B.

Same-input check	Original	Replacement	Speedup
25,001 distinct residual masks	128.91 s	3.91 s	33.00×
Full 12-pair sample, including unchanged setup	246.70 s	117.86 s	2.09×
Requirement-generation stage in that run	see receipt	see receipt	32.77×

Both implementations produce identical minimal-cover sets, all 11 exported requirement records, all 5,509 closure leaves and the complete rank histogram. The full replay regenerates the 25,001-case corpus from 25,991 actual calls. The proof bounds the search tree by 1,023 nodes when (c,d)=(4,5) and establishes minimal-antichain equivalence for arbitrary finite label sets; the timing table is only this machine and sample.

Verification passed: registered Python 3.12 replay; independent explicit-subspace oracle on all 2,048 F2^3 subset/capacity cases; seven rejecting controls; sanitizer-trap kernel replay; known span-five positive outer control; optimized Python 3.14 fast-kernel/oracle replay from a read-only Git-free export; all 19 committed files match tested bytes. No independent human or formal-kernel review is claimed.

python3 -I -B research/two-flat-cover-branching/verify.py --work /tmp/two-flat-replay

Use a fresh work directory, Python 3.12+ and a GNU-C11-capable cc. Expected final line: PASS_TWO_FLAT_COVER_BRANCHING. The registered manifest splits kernel and outer replays into bounded commands.

Scientific base afc158a25ef3e0bd80de81d1eb5339f0dace118e; branch synchronized to a23bdc1b07be96bea0101d44ae27d14de67a79ed before publication. Exact source: DreamingOfClouds' help package at 1b25759bff5486293a9083ebbe0097b5f6642e08. Credit that source's existing dfs4 feasibility routine and #2816/#2826/#2862 lineage. Tekalign with substantial OpenAI Codex assistance, Apache-2.0.

Only research/two-flat-cover-branching/ is added: 19 files, 1,089 added text lines, under 1 MB including compressed inputs. Frozen sources, shared code, upstream/ and the selected frontier are untouched. The complete outer census remains too large; catalogue/orbit completeness, RRC and original integer-rank obligations are not newly discharged. This is a useful research accelerator, not a claim that the class-C problem is solved. Recent κ constructions #2957/#2966/#2969 concern different supplier changes.