| { |
| "columns": [ |
| "Id", |
| "Question", |
| "Design", |
| "Compute (box-hours)", |
| "Success criterion", |
| "Depends on", |
| "Status", |
| "Source" |
| ], |
| "widths": [ |
| 7, |
| 44, |
| 52, |
| 16, |
| 44, |
| 14, |
| 14, |
| 40 |
| ], |
| "rows": [ |
| [ |
| "X0", |
| "Does our stack reproduce Goedel-V2-8B?", |
| "miniF2F-test pass@32 @40k, strict verifier", |
| "~12", |
| "Within noise of 84.6%", |
| "-", |
| "Done: 202/244 (82.8%)", |
| "45f6f99" |
| ], |
| [ |
| "X1", |
| "Pool pass-rate distribution", |
| "2,000 x 8 @16k, overlapped verify", |
| "~25", |
| "Band >= 150 statements", |
| "X0", |
| "Running", |
| "chain4.log" |
| ], |
| [ |
| "G0", |
| "Is the measurement trustworthy?", |
| "Frozen verifier + anchor re-verify; coverage; 4-stage decontam; class-B ledger; RL step fits; Rust pilot", |
| "~30", |
| "All 7 Phase-0 items pass", |
| "X0", |
| "Planned", |
| "SOTA_PLAN.md Phase 0" |
| ], |
| [ |
| "K3", |
| "Feedback vs restarts", |
| "2 repairs/trajectory vs restart sampling, matched tokens, valid", |
| "~15", |
| "Paired valid difference, confirmed with fresh seed", |
| "G0", |
| "Planned", |
| "SOTA_PLAN.md section 3" |
| ], |
| [ |
| "K4", |
| "Averaging alpha", |
| "alpha in {0.6,0.7,0.8,0.9,1.0} after one L3 round, valid", |
| "~15", |
| "Best alpha; selections logged", |
| "G0", |
| "Planned", |
| "SOTA_PLAN.md section 3" |
| ], |
| [ |
| "K2", |
| "Is mutation the active ingredient?", |
| "2,000 verified mutations vs 2,000 extra unmutated proofs, same rows/tokens", |
| "~30", |
| "Winner scaled to 10-15k", |
| "G0", |
| "Planned", |
| "SOTA_PLAN.md section 3" |
| ], |
| [ |
| "K1", |
| "EI-SFT vs RL from the same checkpoint", |
| "10-15 RL steps vs token-matched L3 round; winner continued", |
| "~15-55", |
| "Winner chosen on valid, confirmed n=32 fresh seed", |
| "G0 (RL step)", |
| "Planned", |
| "SOTA_PLAN.md section 3" |
| ], |
| [ |
| "F1", |
| "PRIMARY final", |
| "Class A miniF2F-test, n=k=32, exact McNemar vs anchor", |
| "~23", |
| "~+9 net (~211/244) for significance", |
| "K1-K4", |
| "Planned", |
| "SOTA_PLAN.md section 1" |
| ], |
| [ |
| "F2", |
| "Secondary + exploratory finals", |
| "Class B miniF2F; ProofNet class A (Holm); class C ProofNet; Putnam pass@8", |
| "~160", |
| "Reported with protocol tuple", |
| "F1", |
| "Planned", |
| "SOTA_PLAN.md section 5" |
| ], |
| [ |
| "B1", |
| "Bug study (separate paper)", |
| "Aeneas (Lean 4.31 env), prove/negate closed proposition, vs LLM review vs cargo-fuzz", |
| "~40", |
| "Validated precision per method; executable counterexamples", |
| "G0 pilot", |
| "Planned", |
| "SOTA_PLAN.md Phase 5" |
| ] |
| ] |
| } |
