Verify six-skip/T740 release: conditional kappa 0.001729737950064279

[jeremydiviney]

PR3042's six restore skips retain the ordered helper widths of PR2979 but use the T74 allocation. This package composes those exact literal words with Petalas's PR3024/3030 T740 cap allocation and reruns the complete source/native/owner/physical/prefix/outer release.

The resulting conditional value is κ = 0.001729737950064278551242 − 10^-46 − 10^-70, an exact gain of 178367636256443 / 10^23 (about 1.78e-9) over pinned PR3042. This is a small integration gain from existing constructions. Credit dcharb78 for the six-skip words, Nick Petalas for the allocation and cap, and Marius-Paul Dumitrean for the joint rank-seven predecessor. Full lineage and substantial OpenAI Codex assistance are disclosed in the package.

Concurrent work: a pre-submission refresh found Petalas's same-composition claim, posted during this local run, and #3050's separate code-disjoint substitution/frame audit. No priority or unique composition idea is claimed here. This submission adds a separately executed full release with explicit word/allocation/receipt bindings. It reuses credited scientific engines; separate execution is not code-disjoint verification of those engines.

The complete profile is {1:16376940, 2:4741920, 3:1594737, 4:719280, 5:137085, 8:53280, 48:281903}. The allocation attains its fixed-word packing cap, checked over all 2,551 patterns. Its fixed physical stock increases from 66,813 to 663,549; all 91 cleanup buffers and all 6,660 calls remain charged. The measured prefix requirement is 1,479 bits within the paid 2,048-bit allowance.

Validation covers both exact 123-variable restore matrices, all 632 forward/dual operator columns, each ordered helper width, all 2,717,280 owner roles, the exact complete moment/cap, and semantic rejection controls. The full composed release passed using two fresh current five-stage native chains and explicitly authenticated retained parent/native reuse. The public reproducer defaults to fresh retained bootstrap; that more expensive full mode was not run in this session. The registered replay recomputes finite checks and authenticates stored release evidence; it does not rerun the full native/outer chain.

Reproduce the bounded package check:

python3.12 -B scripts/research_package.py --package research/rank7-six-skip-cap-release

The package README documents the separate full-replay command, exact source commits, payload-based comparison for regenerated gzip inputs, raw receipts, logs, and retained obligations. The authoritative PR3042 literal edits pass; their block lengths/carriers differ from some upstream prose, which this package does not use as proof.

All inherited compiler/all-size, physical backing/movement, precision/tape, caller and analytic assumptions remain. This submission does not change the selected result or claim independent human/formal-kernel review. The useful next step is an independent replay and audit of a named remaining admission implication; further packing-only search at these fixed words and capacities cannot improve the attained cap.

Exact predecessors: #3042 at 84cc406ad91ec53fff329a11a445d7effe564a77; #3030 at 78e77d820393c6b8648b47af219e991edfcc54e6; main base d8e0f83f00191df62174b641e9fff055c357f641 (priority revision 6).