| { |
| "scope": "80 fixed requests, both explicit efforts; merged-model comparison, not independent generalization evidence", |
| "checkpoint_step": 3000, |
| "base_id": "Qwen/Qwen3.5-9B", |
| "base_revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a", |
| "comparison": { |
| "passed": false, |
| "gate": "identical160_argmax_and_max_probability_delta_le_0.05", |
| "depths": { |
| "16": { |
| "before_correct": 63, |
| "after_correct": 63, |
| "changed_decisions": [], |
| "max_probability_abs_difference": 0.08810508251190186, |
| "mean_probability_abs_difference": 0.0019464968157012663, |
| "max_logit_abs_difference": 0.265625, |
| "mean_logit_abs_difference": 0.02963104248046875 |
| }, |
| "32": { |
| "before_correct": 68, |
| "after_correct": 68, |
| "changed_decisions": [], |
| "max_probability_abs_difference": 0.062176525592803955, |
| "mean_probability_abs_difference": 0.001304240933347387, |
| "max_logit_abs_difference": 0.25, |
| "mean_logit_abs_difference": 0.03828125 |
| } |
| }, |
| "before_sha256": "28412c70d8b456b35bb884a2fba4e85801b0135fc7957190927994a56d268a6f", |
| "after_sha256": "3f01a868d4fba91a8a1749df37a87f10f0874823eec5a978291c93231773c365", |
| "verified_at_unix": 1789984423.7235599 |
| }, |
| "runtime_validation": { |
| "status": "passed", |
| "decisions": 160, |
| "max_probability_delta": 0.0, |
| "no_jev_import": true, |
| "long_input_rejected": true, |
| "invalid_effort_rejected": true, |
| "text_scope": "Execution smoke only; not TYPE accuracy or speed validation", |
| "source_sha256": "70d686828090965063966dddac6d8454cf981c6614a406c1647617986e534c5d" |
| }, |
| "decision_release": { |
| "passed": true, |
| "scope": "Independent BF16 merged variant, frozen80 decision parity; NOT a probability-equivalent replacement", |
| "post_observation_scope_amendment": true, |
| "original_numerical_gate_passed": false, |
| "original_gate": "identical160_argmax_and_max_probability_delta_le_0.05", |
| "original_comparison_sha256": "d2ea8a7af10f09fdfde166384ffb18fb5a19ebf1a1281c58cb72806a45878c41", |
| "decision_agreement": 160, |
| "total_decisions": 160, |
| "independent_questions": 80, |
| "probability_calibration_transfer_validated": false, |
| "automatic_routing_validated": false, |
| "required_followup": "Recalibrate all probability, rejection and routing thresholds; independently evaluate new tasks", |
| "depths": { |
| "16": { |
| "before_correct": 63, |
| "after_correct": 63, |
| "changed_decisions": [], |
| "max_probability_abs_difference": 0.08810508251190186, |
| "mean_probability_abs_difference": 0.0019464968157012663, |
| "max_logit_abs_difference": 0.265625, |
| "mean_logit_abs_difference": 0.02963104248046875 |
| }, |
| "32": { |
| "before_correct": 68, |
| "after_correct": 68, |
| "changed_decisions": [], |
| "max_probability_abs_difference": 0.062176525592803955, |
| "mean_probability_abs_difference": 0.001304240933347387, |
| "max_logit_abs_difference": 0.25, |
| "mean_logit_abs_difference": 0.03828125 |
| } |
| }, |
| "publication_requires": "Further portable-runtime160, weight arithmetic, full file hashes and fresh HF reload gates" |
| }, |
| "reviewed_variant": {}, |
| "decision_parity_passed": true, |
| "decision_agreement": 160, |
| "weight_arithmetic_status": "passed", |
| "raw_inputs_included": false |
| } |
|
|
