| { |
| "arm_mean_acceptance": { |
| "run1_fim": 0.7848643666505813, |
| "run1_fim_shuf": 0.7643998256325721, |
| "run1_l2r": 0.779159582555294, |
| "run2_fim": 0.6985759252309799, |
| "run2_fim_shuf": 0.683045190423727, |
| "run2_l2r": 0.7791734331846237 |
| }, |
| "baseline_revision_evidence": { |
| "kind": "local_huggingface_cache_refs_captured_after_run1_corpus_build", |
| "limitation": "This is repository-revision evidence, not a per-document raw-corpus manifest for run 1. Source drift between the run 1 build and the captured cache refs cannot be ruled out." |
| }, |
| "depth": 2, |
| "documents": 200, |
| "inputs": { |
| "ablation": { |
| "path": "out/run2-no-fim/acceptance.no-fim-ablation.json", |
| "sha256": "9a29fc1346b77fd3703e6b502f57276b9cf22aee904487a9ea2beac5e85a8681" |
| }, |
| "baseline": { |
| "path": "out/run1/acceptance.trained.json", |
| "sha256": "53214c7cc098a3d29548563730855e16876ae28a615d61db62002dc170d81b0e" |
| }, |
| "format_receipt": { |
| "path": "config/eval_format_ablation_receipt.json", |
| "sha256": "827ecc920fa6dd4cfe2922fde4d2289967c9e32759f368b466761302cf27b3c0" |
| }, |
| "source_receipt": { |
| "path": "config/eval_holdout_receipt.json", |
| "sha256": "d24d4d8e41d75c91b2ff72a04f54a4519c15f1b362e70664cbfe1a12c736601d" |
| } |
| }, |
| "instrument_version": 4, |
| "primary_endpoint": { |
| "ci95": [ |
| 0.07469401128590107, |
| 0.0882960123345256 |
| ], |
| "contract": { |
| "bootstrap_samples": 4000, |
| "bootstrap_seed": 0, |
| "bootstrap_unit": "paired_target_document", |
| "depth": 2, |
| "metric": "(run1_fim_shuf_minus_l2r)_minus_(run2_fim_shuf_minus_l2r)", |
| "positive_when": "ci95_lower_gt_0" |
| }, |
| "difference_in_differences": 0.08136848583817483, |
| "run1_fim_shuf_minus_l2r": -0.014759756922721912, |
| "run2_fim_shuf_minus_l2r": -0.09612824276089671, |
| "verdict": "POSITIVE" |
| }, |
| "publication_ready": true, |
| "research_outcomes_are_not_release_gates": true, |
| "runtime_code_evidence": { |
| "kind": "matched_sampler_reset_with_post_build_shard_attestation", |
| "limitation": "Run 1 did not record source-file hashes or sampler state in its pre-fix checkpoints. A post-build receipt now hashes all 52 current shard files and binds deterministic visible-grammar normalization, but it does not prove that the bytes were unchanged since training began or recover exact original units. The step-300 process recovery reset the legacy sampler, causing 78,643,200 scheduled token positions, 1.57 percent of the training budget, to replay earlier random windows. sampler_reset_steps [300] matches that reset timing in both arms. Later checkpoints preserve exact sampler RNG state. Exact sampled-example equivalence still cannot be proven." |
| }, |
| "schema_version": 1, |
| "secondary_endpoint": { |
| "ci95": [ |
| -0.0025777602791786196, |
| 0.012247835688292982 |
| ], |
| "contract": { |
| "bootstrap_samples": 4000, |
| "bootstrap_seed": 0, |
| "bootstrap_unit": "paired_target_document", |
| "depth": 2, |
| "metric": "(run1_fim_minus_fim_shuf)_minus_(run2_fim_minus_fim_shuf)" |
| }, |
| "difference_in_differences": 0.004933806210756302, |
| "run1_fim_minus_fim_shuf": 0.020464541018009208, |
| "run2_fim_minus_fim_shuf": 0.015530734807252866, |
| "verdict": "NULL: the interval includes zero" |
| } |
| } |
|
|
