Repository navigation
Conditional κ = 0.00160109005167629: PR568 kernels and eight-cold provider on PR571's prefix word - #576
Conversation
…provider on PR571's prefix word
…e three-provider witness
f4e9f7e to
269490b
Compare
|
Thank you @jeremykothe2. This contribution at Adds twelve multi-cut kernels to the #571 prefix as a third real provider. The final spec supplies expected comparisons, not a replacement for reconstruction. Both changed providers and the mixed allocation must pass. Preserved implementation and research: The family review and individual disposition record the scope, evidence and attribution. Parallel and intermediate contributions remain credited even where a later composition has a stronger bound. The published multiplication bound remains conditional on the documented compiler, realization, hosting and analytic assumptions; these finite checks are not a full Lean multiplication proof. |
Summary
Conditional κ = 0.00160109005167629. This is +4.08e-7 over PR571 (0.001600682403612293), +3.20e-7 over PR575 (0.001600770286423653) and +2.42e-7 over PR577 (0.001600848528154892). The conservative model gives 0.001601090048890878.
The construction keeps DominikScholz's PR571 joint-prefix word and adds three providers:
A new literal owner map hosts all 72 calls. The paid profile is
{1:169688,2:40193,3:26386,4:13824,5:4032,6:4032,8:576,48:1817}.How it was found
build_multicut_kernels.py, with an exact zero dirty response.Package
research/complex-571-three-provider/contains 16 files. PR568 (research/complex-multicold,71c99ac) and PR571 (research/structural-gain,b7bdc5a) are pinned external sources, and nothing is copied from them.The global stage is driven by data.
code/reads every expected value fromdata/WITNESS-SPEC.json: providers, dispatch, the owner and plan hashes, the profile and κ. It recomputes each value from the words, owners and engines and compares it with the spec. No witness constants appear in code. The audit and assembly mutation controls are generated per provider: 31 owner mutations and 147 assembly controls, each of which must be rejected.Verification
PR568 and PR571 are not merged, so fetch their pinned commits first:
verify.pyuses only the standard library. It runs:Local result (Windows, CPython 3.11.11):
PASS_PR571_PREFIX_WORD_PREFIXK_AND_PR568_EIGHT_COLD_THREE_PROVIDER_NATIVE_COMPLEX_AND_CITED_BIT, κ 0.00160109005167629, 31/31 stages exit 0, 348 s. The verified package tree is byte-identical to this commit.The first commit on this PR (κ 0.001600951400768614) also passed 31/31 (347 s). The second commit changes only
data/PLAN.json,data/OWNERS.json.gz,data/WITNESS-SPEC.jsonandMANIFEST.jsonto an improved literal owner map; providers, dispatch and code are unchanged.GitHub CI has not run, and I make no claim about it.
Credits
AI disclosure and authorship: Claude Opus 5.5 (Anthropic), running in the omp coding harness, did the work for this PR: the kernel search, the owner solve, the spec-driven package and this description. Jeremy Kothe (jeremykothe2) directed and reviewed it and is submitting it.
Scope
The result is conditional on the inherited all-size compiler, common-label/mutable-source realization, whole-call hosting, precision/tape/exact recovery, analytic transfer and the bit H1–H5 assumptions. It is finite exact evidence. It is not an unconditional theorem, a complete Lean proof, or a practical speedup.