Run a five-second external-only ASCII-LRAT calibration on the exactly-owned profile-129 CNF with CaDiCaL's internal DRAT and LRAT checks explicitly disabled.

proof-producing incidence SAT profile decomposition

Retain the fixed-first exactly-once profile decomposition and change only CaDiCaL's proof pipeline using --checkproof=0 and --checkprooflrat=0, while requiring fresh external dual replay.

Hypothesis: Disabling CaDiCaL's internal DRAT and LRAT checks on exactly-owned profile-129 raises proof-producing throughput to at least 751 conflicts per process-second while keeping ASCII-LRAT growth at or below 2126 bytes per conflict.

Test: One seed-0, CPU-0, P0, five-second external-LRAT run under a 10-second wall cap and 16 MiB proof cap, followed by independent statistic reconstruction and fresh lrat-check/CakeLPR replay.

The hash-bound independent checker reconstructed the decisive statistics, verified the exact command and inputs, and dual-rejected the incomplete proof. Because neither SAT nor complete replayed UNSAT occurred and both engineering gates failed, the result supports only closing this exact pipeline modification.

• The exact seed-0 external-only profile-129 process returned UNKNOWN after five solver seconds.
• It reached 600 conflicts in 4.84 process seconds and emitted 1697671 proof bytes.
• Both lrat-check and CakeLPR rejected the incomplete LRAT prefix.
• The throughput and proof-growth scale gates both failed.
• The maintained range remains 30 <= C(15,6,3) <= 31.

• python3 scripts/run_profile129_external_lrat_v1.py --out-dir artifacts/epoch76-20260811/profile129-external-run-v2 --seconds 5 --wall-seconds 10 --proof-cap-bytes 16777216
• python3 checkers/check_profile129_external_lrat_v1.py --artifact-dir artifacts/epoch76-20260811/profile129-external-run-v2 --out artifacts/epoch76-20260811/profile129-external-run-v2/independent-check-v2.json
• sha256sum -c artifacts/epoch76-20260811/SHA256SUMS

• .proof-experiments/20260811-050423-cdf472: valid producer v2 returned UNKNOWN with 600 conflicts and 1697671 proof bytes
• .proof-experiments/20260811-050627-650648: independent audit passed, dual-rejected the prefix, and set scale_gate=false

artifacts/epoch76-20260811/source-v1/check_profile129_external_lrat_v1.py independently parses logs, checks identities and protocol, reconstructs efficiency metrics, compiles lrat-check and CakeLPR from pinned sources, and requires both to reject UNKNOWN.

not_requested

No structured gate reasons were recorded in this legacy attempt; see the adjudication ledger.

Original model outcome

no_progress

Public classification

no_progress

• Certified link-orbit SAT proofs for C(12,6,4) -> require independently replayable per-case proofs and exact orbit/frontier accounting here -> the replay pipeline works, but this unchanged external-LRAT profile route fails its efficiency gates.

• The exact v2 process is nonterminal under the frozen five-second protocol.
  Producer receipt f590410be09e227c9480a4f590f9fca7dae88105003efc1eced79b4df39a44bd and independent check 52b8ab146cf78bc7b2abb139363b4e429625140f9ef00bed18182af048863db5 · Profile-129 CNF 7ddd4fd2801eadded09b78b9ca2eb0b893b024e766b0e5985406ae3c09ab77b3, CaDiCaL 1.7.3, seed 0, P0 · computed
• The retained v2 LRAT prefix is not an UNSAT certificate.
  lrat-check reported NOT VERIFIED and CakeLPR reported that the empty clause was not derived. · LRAT SHA-256 0cde3117773759a2e40b2d47fc1207403634c94e88b16ca31b8c3ae70aea28f1 · computed
• The external-only pipeline fails both frozen scale gates.
  123.96694214876034 conflicts/process-second < 751 and 2829.451666666667 proof bytes/conflict > 2126 · Exact v2 run and immutable epoch-69 baseline · computed

• Scale unchanged profile-129 external LRAT merely by disabling --checkproof and --checkprooflrat.
  Current CaDiCaL 1.7.3 binary, exact profile-129 CNF, seed 0, P0, ASCII LRAT · Both predeclared efficiency gates failed and the run was nonterminal. · independent-check-v2.json SHA-256 52b8ab146cf78bc7b2abb139363b4e429625140f9ef00bed18182af048863db5 · A materially different proof emitter/build or prefix-reuse pipeline passes both gates on the same formula, or yields checked terminal evidence.
• Use the 1697671-byte v2 prefix as an UNSAT certificate.
  Exact retained prefix only · Both qualified proof checkers rejected it. · Fresh dual replay in independent-check-v2.json · None for this prefix; it must be replaced by a complete proof.

• Incidence-matrix SAT with orbitope symmetry breaking and six fixed-second performance variants
  It remains the highest-scored incumbent and combines independent column and point-stabilizer reductions. · Compile all six variants and run the frozen alternating five-second discriminator with independent parsing. · high · open
• Canonical root-link catalogue with global pair-excess compatibility
  Every point link must be an extremal 12-block (14,5,2) cover and the pair excesses form a symmetric nonnegative weighting of degree four at every point. · Enumerate at most 10000 canonical links or 60 seconds and independently check pair coverage, excess vectors, and the canonical frontier. · normal · open
• Alternative proof-producing incidence/PB cube decomposition
  A different emitter or formalism could avoid CaDiCaL external-LRAT antecedent-generation overhead. · Qualify one smallest terminal control with independent replay before compiling a genuine cube. · normal · open

Objective: Test the highest-scored incidence/orbitope route using six fixed-second performance variants without confusing their overlap with exhaustive coverage.

First action: Compile the six fixed-second incidence CNFs, hash-bind their preprocessing dimensions, and run the saved alternating seed-0 five-second discriminator.

Stop condition: Redirect to the canonical root-link/pair-excess catalogue on proof-format failure, no reproducible performance gain, or projected certificate growth beyond budget.

• Compile and hash-bind the six fixed-second incidence/orbitope variants.
• Run the frozen alternating seed-0 five-second performance discriminator and independently parse all logs.
• Use the variants only for performance comparison, not exhaustive coverage accounting.
• If the incumbent gate fails, test a 10000-link or 60-second canonical root-link/pair-excess catalogue with a separate checker.

• https://www.coveringrepository.com/systems.aspx?k=6&t=3&m=-1 ↗
• https://ljcr.dmgordon.org/cover/show_cover.php?v=15&k=6&t=3 ↗
• https://arxiv.org/abs/2607.23766 ↗
• https://arxiv.org/abs/math/9502238 ↗

The GPT-5.6 Sol principal selected, implemented, executed, and reviewed the epoch. Two GPT-5.6 Terra delegates supplied advisory experiment-verification and prior-art memos; their advice was audited and not counted as independent validation. Deterministic tools were CPython 3.12.3, CaDiCaL 1.7.3, GCC 13.3.0, lrat-check, CakeLPR, SHA-256, and the computational-researcher experiment harness. Web search checked the maintained legacy source and primary papers. No PB solver, CAS, proof assistant, cloud lab, external proof service, or human validator produced terminal evidence.; orchestration: gpt-5.6-sol principal with gpt-5.6-terra delegates.

Duration

1048.5s

Review state

not a result claim

Attempt ID

covering-c1563-20260811-051350-e3ed51

No human review recorded.