laya-evaluation / banking_execution_audit.json
c41's picture
Publish evaluation preview and verified historical classification tables
ff51f29 verified
Raw History Blame Contribute Delete
1.01 kB
{
"status": "completed_with_unresolved_exact_reproduction",
"historical_accuracy_percent": 94.09090909,
"eager_fixed_weight_accuracy_percent": 93.99350649,
"original_standard_script_current_accuracy_percent": 94.05844156,
"historical_correct": 2898,
"eager_correct": 2895,
"original_script_current_correct": 2897,
"n": 3080,
"historical_exact_prediction_match": false,
"original_script_different_predictions": 3,
"data_sha256_matches_historical": true,
"weights_sha256_matches_historical": true,
"eight_eager_conditions_correct": 2895,
"cause_scope": "Execution-path sensitivity demonstrated. Precision/batch/wrapper alone with compilation disabled do not recover the historical count. Single operator and remaining historical drift not isolated.",
"continuation_best_fp32_percent": 94.12337662,
"continuation_sam_mean_fp32_percent": 94.03679654,
"significant_gain_demonstrated": false,
"public_scores_changed": false,
"no_test_driven_tuning": true
}