| { |
| "preset": "quick", |
| "rows": 477, |
| "correct": 309, |
| "micro_accuracy": 0.6477987421383647, |
| "jevbench_correct": 135, |
| "jevbench_rows": 231, |
| "semif_correct": 174, |
| "semif_rows": 246, |
| "format": "laya_native", |
| "probabilities": "raw decision softmax; no fitted calibration", |
| "suites": { |
| "jevbench-original": { |
| "rows": 72, |
| "successful_rows": 72, |
| "failed_rows": 0, |
| "accuracy": 0.625, |
| "equal_case_modal_agreement": 0.625, |
| "mean_family_balanced_accuracy": 0.6319444444444444, |
| "equal_case_total_variation_successful_only": null, |
| "total_variation_rows": 0, |
| "dataset_sha256": "11f3f54e9ead94337635573856061aae4e9c7e25147147a25767d79a20cffe90", |
| "adapter_sha256": "9f1ab94e0fe312fcbb4c0872260fac051fbe27430e38e2723f4924a1454eabf8" |
| }, |
| "jevbench-easy": { |
| "rows": 48, |
| "successful_rows": 48, |
| "failed_rows": 0, |
| "accuracy": 0.9791666666666666, |
| "equal_case_modal_agreement": 0.9791666666666666, |
| "mean_family_balanced_accuracy": 0.9791666666666667, |
| "equal_case_total_variation_successful_only": null, |
| "total_variation_rows": 0, |
| "dataset_sha256": "c90f37652798af7630c46ee35beb514b74312f1eb5d5e9b03fa7205a7d682b8c", |
| "adapter_sha256": "9f1ab94e0fe312fcbb4c0872260fac051fbe27430e38e2723f4924a1454eabf8" |
| }, |
| "jevbench-hard": { |
| "rows": 111, |
| "successful_rows": 111, |
| "failed_rows": 0, |
| "accuracy": 0.38738738738738737, |
| "equal_case_modal_agreement": 0.38738738738738737, |
| "mean_family_balanced_accuracy": 0.4583134920634921, |
| "equal_case_total_variation_successful_only": null, |
| "total_variation_rows": 0, |
| "dataset_sha256": "edfda79bfa327392e90da818ed38dcdbbf31b9755b8a381d3aca4423d11f466c", |
| "adapter_sha256": "9f1ab94e0fe312fcbb4c0872260fac051fbe27430e38e2723f4924a1454eabf8" |
| }, |
| "semif-authored": { |
| "rows": 144, |
| "successful_rows": 144, |
| "failed_rows": 0, |
| "accuracy": 0.7777777777777778, |
| "equal_case_modal_agreement": 0.7777777777777778, |
| "mean_family_balanced_accuracy": 0.7933897243107769, |
| "equal_case_total_variation_successful_only": null, |
| "total_variation_rows": 0, |
| "dataset_sha256": "8162d1c73f925af64453f1ec05ef36d583b3815bf698e60f0d454bd11537e079", |
| "adapter_sha256": "c088c9277435df01845009b36e270731754a1558748a84b990ea7c3e11c2fe8c" |
| }, |
| "semif-typesafe": { |
| "rows": 102, |
| "successful_rows": 102, |
| "failed_rows": 0, |
| "accuracy": 0.6078431372549019, |
| "equal_case_modal_agreement": 0.6436721611721612, |
| "mean_family_balanced_accuracy": 0.4726731601731602, |
| "equal_case_total_variation_successful_only": 0.37759414352433773, |
| "total_variation_rows": 102, |
| "dataset_sha256": "734bfa7c56a1e4616e3dee56a71b3dd44b2717def0644622f00c6006d3e80f75", |
| "adapter_sha256": "c088c9277435df01845009b36e270731754a1558748a84b990ea7c3e11c2fe8c" |
| } |
| } |
| } |
