File size: 84,027 Bytes
d053378 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 | {
"device": "cuda",
"eval": "data/tmp/mixed_mode_eval.v2.field_decision.jsonl",
"eval_candidate_pairs": 21986,
"eval_groups": 1900,
"eval_rows": 220,
"last_loss": 0.08536958694458008,
"max_steps": 160,
"model": "answerdotai/ModernBERT-base",
"output_dir": "models/modernbert-field-event-ranker-mixed-v2-perf-H4",
"peak_allocated_gib": 4.076,
"peak_reserved_gib": 4.17,
"positive_train_pairs": 12854,
"skip_save_model": false,
"source_rank": {
"mean_expected_source_rank": 3.0918580375782883,
"mean_reciprocal_rank": 0.5255961949014605,
"median_expected_source_rank": 2,
"missing_prediction_pairs": 0,
"per_field": {
"action_causality": {
"mean_rank": 3.00709219858156,
"mean_reciprocal_rank": 0.48304626815265134,
"missing_prediction_pairs": 0,
"ranked_pairs": 141,
"source_pairs": 141,
"top_k_hits": {
"1": 34,
"2": 61,
"3": 92,
"5": 127
},
"top_k_recall": {
"1": 0.24113475177304963,
"2": 0.4326241134751773,
"3": 0.6524822695035462,
"5": 0.900709219858156
}
},
"assistant_claims_to_verify": {
"mean_rank": 2.0,
"mean_reciprocal_rank": 0.5,
"missing_prediction_pairs": 0,
"ranked_pairs": 1,
"source_pairs": 1,
"top_k_hits": {
"1": 0,
"2": 1,
"3": 1,
"5": 1
},
"top_k_recall": {
"1": 0.0,
"2": 1.0,
"3": 1.0,
"5": 1.0
}
},
"attempt_outcome_pairs": {
"mean_rank": 2.3773584905660377,
"mean_reciprocal_rank": 0.5625786163522012,
"missing_prediction_pairs": 0,
"ranked_pairs": 53,
"source_pairs": 53,
"top_k_hits": {
"1": 16,
"2": 31,
"3": 43,
"5": 52
},
"top_k_recall": {
"1": 0.3018867924528302,
"2": 0.5849056603773585,
"3": 0.8113207547169812,
"5": 0.9811320754716981
}
},
"attempted_actions": {
"mean_rank": 3.756501182033097,
"mean_reciprocal_rank": 0.4687886557994194,
"missing_prediction_pairs": 0,
"ranked_pairs": 846,
"source_pairs": 846,
"top_k_hits": {
"1": 196,
"2": 383,
"3": 531,
"5": 713
},
"top_k_recall": {
"1": 0.23167848699763594,
"2": 0.45271867612293143,
"3": 0.6276595744680851,
"5": 0.8427895981087471
}
},
"customer_identity": {
"mean_rank": 1.5454545454545454,
"mean_reciprocal_rank": 0.8181818181818182,
"missing_prediction_pairs": 0,
"ranked_pairs": 11,
"source_pairs": 11,
"top_k_hits": {
"1": 8,
"2": 8,
"3": 11,
"5": 11
},
"top_k_recall": {
"1": 0.7272727272727273,
"2": 0.7272727272727273,
"3": 1.0,
"5": 1.0
}
},
"discarded_options": {
"mean_rank": 3.0,
"mean_reciprocal_rank": 0.375,
"missing_prediction_pairs": 0,
"ranked_pairs": 2,
"source_pairs": 2,
"top_k_hits": {
"1": 0,
"2": 1,
"3": 1,
"5": 2
},
"top_k_recall": {
"1": 0.0,
"2": 0.5,
"3": 0.5,
"5": 1.0
}
},
"explicit_decisions": {
"mean_rank": 1.0,
"mean_reciprocal_rank": 1.0,
"missing_prediction_pairs": 0,
"ranked_pairs": 1,
"source_pairs": 1,
"top_k_hits": {
"1": 1,
"2": 1,
"3": 1,
"5": 1
},
"top_k_recall": {
"1": 1.0,
"2": 1.0,
"3": 1.0,
"5": 1.0
}
},
"failed_attempts": {
"mean_rank": 2.5,
"mean_reciprocal_rank": 0.6150793650793651,
"missing_prediction_pairs": 0,
"ranked_pairs": 36,
"source_pairs": 36,
"top_k_hits": {
"1": 15,
"2": 22,
"3": 28,
"5": 34
},
"top_k_recall": {
"1": 0.4166666666666667,
"2": 0.6111111111111112,
"3": 0.7777777777777778,
"5": 0.9444444444444444
}
},
"initiating_command": {
"mean_rank": 2.5728643216080402,
"mean_reciprocal_rank": 0.667125852929873,
"missing_prediction_pairs": 0,
"ranked_pairs": 199,
"source_pairs": 199,
"top_k_hits": {
"1": 99,
"2": 139,
"3": 160,
"5": 178
},
"top_k_recall": {
"1": 0.49748743718592964,
"2": 0.6984924623115578,
"3": 0.8040201005025126,
"5": 0.8944723618090452
}
},
"invalidation_hints": {
"mean_rank": 1.5,
"mean_reciprocal_rank": 0.75,
"missing_prediction_pairs": 0,
"ranked_pairs": 2,
"source_pairs": 2,
"top_k_hits": {
"1": 1,
"2": 2,
"3": 2,
"5": 2
},
"top_k_recall": {
"1": 0.5,
"2": 1.0,
"3": 1.0,
"5": 1.0
}
},
"next_actions": {
"mean_rank": 2.875,
"mean_reciprocal_rank": 0.5649440836940836,
"missing_prediction_pairs": 0,
"ranked_pairs": 32,
"source_pairs": 32,
"top_k_hits": {
"1": 11,
"2": 19,
"3": 23,
"5": 29
},
"top_k_recall": {
"1": 0.34375,
"2": 0.59375,
"3": 0.71875,
"5": 0.90625
}
},
"non_promotable_context": {
"mean_rank": 3.0,
"mean_reciprocal_rank": 0.3333333333333333,
"missing_prediction_pairs": 0,
"ranked_pairs": 2,
"source_pairs": 2,
"top_k_hits": {
"1": 0,
"2": 0,
"3": 2,
"5": 2
},
"top_k_recall": {
"1": 0.0,
"2": 0.0,
"3": 1.0,
"5": 1.0
}
},
"observed_outcomes": {
"mean_rank": 3.0272189349112426,
"mean_reciprocal_rank": 0.4855734009580153,
"missing_prediction_pairs": 0,
"ranked_pairs": 845,
"source_pairs": 845,
"top_k_hits": {
"1": 199,
"2": 393,
"3": 557,
"5": 754
},
"top_k_recall": {
"1": 0.23550295857988165,
"2": 0.4650887573964497,
"3": 0.659171597633136,
"5": 0.8923076923076924
}
},
"outcome_of_latest_attempt": {
"mean_rank": 1.6685082872928176,
"mean_reciprocal_rank": 0.7629439621152327,
"missing_prediction_pairs": 0,
"ranked_pairs": 181,
"source_pairs": 181,
"top_k_hits": {
"1": 108,
"2": 149,
"3": 171,
"5": 179
},
"top_k_recall": {
"1": 0.5966850828729282,
"2": 0.8232044198895028,
"3": 0.9447513812154696,
"5": 0.988950276243094
}
},
"payment_or_warranty_detail": {
"mean_rank": 3.5,
"mean_reciprocal_rank": 0.35,
"missing_prediction_pairs": 0,
"ranked_pairs": 2,
"source_pairs": 2,
"top_k_hits": {
"1": 0,
"2": 1,
"3": 1,
"5": 2
},
"top_k_recall": {
"1": 0.0,
"2": 0.5,
"3": 0.5,
"5": 1.0
}
},
"product_name": {
"mean_rank": 1.0,
"mean_reciprocal_rank": 1.0,
"missing_prediction_pairs": 0,
"ranked_pairs": 4,
"source_pairs": 4,
"top_k_hits": {
"1": 4,
"2": 4,
"3": 4,
"5": 4
},
"top_k_recall": {
"1": 1.0,
"2": 1.0,
"3": 1.0,
"5": 1.0
}
},
"recent_error": {
"mean_rank": 1.8620689655172413,
"mean_reciprocal_rank": 0.7425287356321838,
"missing_prediction_pairs": 0,
"ranked_pairs": 29,
"source_pairs": 29,
"top_k_hits": {
"1": 17,
"2": 23,
"3": 25,
"5": 28
},
"top_k_recall": {
"1": 0.5862068965517241,
"2": 0.7931034482758621,
"3": 0.8620689655172413,
"5": 0.9655172413793104
}
},
"resolved_context": {
"mean_rank": 4.0,
"mean_reciprocal_rank": 0.25,
"missing_prediction_pairs": 0,
"ranked_pairs": 1,
"source_pairs": 1,
"top_k_hits": {
"1": 0,
"2": 0,
"3": 0,
"5": 1
},
"top_k_recall": {
"1": 0.0,
"2": 0.0,
"3": 0.0,
"5": 1.0
}
},
"touched_files": {
"mean_rank": 3.0,
"mean_reciprocal_rank": 0.3333333333333333,
"missing_prediction_pairs": 0,
"ranked_pairs": 1,
"source_pairs": 1,
"top_k_hits": {
"1": 0,
"2": 0,
"3": 1,
"5": 1
},
"top_k_recall": {
"1": 0.0,
"2": 0.0,
"3": 1.0,
"5": 1.0
}
},
"transaction_reference": {
"mean_rank": 2.6,
"mean_reciprocal_rank": 0.4666666666666666,
"missing_prediction_pairs": 0,
"ranked_pairs": 5,
"source_pairs": 5,
"top_k_hits": {
"1": 1,
"2": 1,
"3": 5,
"5": 5
},
"top_k_recall": {
"1": 0.2,
"2": 0.2,
"3": 1.0,
"5": 1.0
}
},
"unsupported_hypotheses": {
"mean_rank": 3.0,
"mean_reciprocal_rank": 0.3333333333333333,
"missing_prediction_pairs": 0,
"ranked_pairs": 1,
"source_pairs": 1,
"top_k_hits": {
"1": 0,
"2": 0,
"3": 1,
"5": 1
},
"top_k_recall": {
"1": 0.0,
"2": 0.0,
"3": 1.0,
"5": 1.0
}
}
},
"rows": 220,
"selected_fields": 997,
"source_pairs": 2395,
"top_k_hits": {
"1": 710,
"2": 1239,
"3": 1660,
"5": 2127
},
"top_k_source_pair_recall": {
"1": 0.2964509394572025,
"2": 0.5173277661795407,
"3": 0.6931106471816284,
"5": 0.8881002087682672
},
"worst_expected_source_ranks": [
{
"expected_event": {
"command": "python - <<'PY'\nimport json\nfrom pathlib import Path\nfor model in ['qwen3.6-35b-a3b-nvfp4','qwen3.6-27b-text-nvfp4-mtp']:\n p=Path('/home/avifenesh/projects/aviary/models')/model\n print('MODEL', p)\n for f in ['config.json','quantization_config.json','quantize_config.json','generation_config.json']:\n q=p/f\n if q.exists():\n data=json.loads(q.read_text())\n print(f, {k:data.get(k) for k in ['model_type','architectures','quantization_config','kv_lora_rank','num_hidden_layers','num_attention_heads','num_key_value_heads','max_position_embeddings','sliding_window','use_mtp','num_nextn_predict_layers'] if k in data})\nPY",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python - <<'PY'\nimport json\nfrom pathlib import Path\nfor model in ['qwen3.6-35b-a3b-nvfp4','qwen3.6-27b-text-nvfp4-mtp']:\n p=Path('/home/avifenesh/projects/aviary/models')/model...",
"tool_name": "exec_command"
},
"expected_probability": 0.28537195920944214,
"expected_source_event_id": "t6",
"field": "attempted_actions",
"rank": 20,
"reciprocal_rank": 0.05,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w14",
"top_candidates": [
{
"event": {
"command": "{\"session_id\": 92479, \"chars\": \"\\u0003\", \"yield_time_ms\": 1000, \"max_output_tokens\": 20000}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "{\"session_id\": 92479, \"chars\": \"\\u0003\", \"yield_time_ms\": 1000, \"max_output_tokens\": 20000}",
"tool_name": "write_stdin"
},
"label": 1,
"source_event_id": "t3",
"support_probability": 0.9996079802513123
},
{
"event": {
"command": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 30000}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 30000}",
"tool_name": "write_stdin"
},
"label": 1,
"source_event_id": "t12",
"support_probability": 0.9992678761482239
},
{
"event": {
"command": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 6000}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 6000}",
"tool_name": "write_stdin"
},
"label": 1,
"source_event_id": "t17",
"support_probability": 0.9992678761482239
},
{
"event": {
"command": "nvidia-smi --query-gpu=memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp sport = :8000",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "nvidia-smi --query-gpu=memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp sport = :8000",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t13",
"support_probability": 0.9985895752906799
},
{
"event": {
"command": "nvidia-smi --query-gpu=name,memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp 'sport = :8000' || true",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "nvidia-smi --query-gpu=name,memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp 'sport = :8000' || true",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t18",
"support_probability": 0.998245358467102
}
]
},
{
"expected_event": {
"command": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\nroot=Path('.')\ncanonical=set()\nfor p in Path('data/evals').glob('*/scorecards.yml'):\n d=yaml.safe_load(p.read_text()) or {}\n for c in d.get('scorecards') or []:\n if isinstance(c,dict) and c.get('result_path'):\n canonical.add(str(Path(c['result_path']).resolve()))\nprint('canonical result paths',len(canonical))\nnotcanon=[]; incanon=[]\nfor p in Path('data/runs').rglob('judge-scorecard-*.yml'):\n d=yaml.safe_load(p.read_text()) or {}\n rp=d.get('result_path')\n if not rp:\n # common default result.md in same dir\n cand=p.parent/'result.md'\n if cand.exists(): rp=str(cand)\n key=str(Path(rp).resolve()) if rp else ''\n (incanon if key in canonical else notcanon).append((p,key))\nprint('judge ",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\nroot=Path('.')\ncanonical=set()\nfor p in Path('data/evals').glob('*/scorecards.yml'):\n d=yaml.safe_load(p.read_text()) or...",
"tool_name": "exec_command"
},
"expected_probability": 0.4863315224647522,
"expected_source_event_id": "t17",
"field": "attempted_actions",
"rank": 20,
"reciprocal_rank": 0.05,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w48",
"top_candidates": [
{
"event": {
"command": "find data/runs -name 'codex-scorecards-draft.yml' -print | wc -l && find data/runs -name 'codex-scorecards-draft.yml' -print",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "find data/runs -name 'codex-scorecards-draft.yml' -print | wc -l && find data/runs -name 'codex-scorecards-draft.yml' -print",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t4",
"support_probability": 0.999783456325531
},
{
"event": {
"command": "sed -n '1,240p' data/evals/judge-review-queues/20260504-first10.txt",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "sed -n '1,240p' data/evals/judge-review-queues/20260504-first10.txt",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t12",
"support_probability": 0.9996646642684937
},
{
"event": {
"command": "find data/evals/judge-review-queues -type f -maxdepth 4 -print 2>/dev/null | sort && printf '\\nreview markers\\n' && rg -n \"review_status|avi_corrected|pending|draft|accepted|rejected\" data/runs data/evals/judge-review-queues -g 'codex-scorecards-draft.yml' -g '*.yml'",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "find data/evals/judge-review-queues -type f -maxdepth 4 -print 2>/dev/null | sort && printf '\\nreview markers\\n' && rg -n \"review_status|avi_corrected|pending|draft|accepted|rej...",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t5",
"support_probability": 0.998245358467102
},
{
"event": {
"command": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\nfor p in sorted(Path('data/runs').rglob('codex-scorecards-draft.yml')):\n d=yaml.safe_load(p.read_text()) or {}\n cards=d.get('scorecards') or []\n print(f'{p}\\tfile_status={d.get(\"review_status\")}\\tcards={len(cards)}\\tcard_statuses={sorted({str(c.get(\"review_status\", d.get(\"review_status\", \"\"))) for c in cards if isinstance(c,dict)})}')\nPY",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\nfor p in sorted(Path('data/runs').rglob('codex-scorecards-draft.yml')):\n d=yaml.safe_load(p.read_text()) or {}\n cards...",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t11",
"support_probability": 0.9921841025352478
},
{
"event": {
"command": null,
"is_context_tool_echo": false,
"role": "tool_result",
"success": true,
"text": "Chunk ID: c00386\nWall time: 0.0000 seconds\nProcess exited with code 0\nOriginal token count: 499\nOutput:\ncanonical result paths 46\njudge draft files already canonical by result_p...",
"tool_name": null
},
"label": 0,
"source_event_id": "r19",
"support_probability": 0.9759464263916016
}
]
},
{
"expected_event": {
"command": "*** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/data/evals/judge-review-status.yml\n@@\n - result_path: /home/avifenesh/projects/aviary/data/runs/session_summarize_for_avi/20260504-memory-gym-variety/qwen36-27b-autoround-64k/round-13-agent-sh-multirepo-maintenance/result.md\n- status: pending_avi_judge_review\n- note: Recent judge-draft batch; keep waiting for Avi unless verbal correction already exists.\n+ status: reviewed_with_known_owner_correction\n+ reviewed_at: '2026-05-04'\n+ note: Avi correction is score 2; the judges overranked a mistaken, failed, not-helpful session as positive.\n - result_path: /home/avifenesh/projects/aviary/data/runs/session_summarize_for_avi/20260504-memory-gym-variety/qwen36-27b-autoround-64k/round-14-agnix-next-task-workflow-policy/result.md\n- status: pending_avi_judge_review\n- note: Recent judge-draft batch; keep waiting for Avi unless verbal correction already exists.\n+ status: reviewed_with_known_owner_correction\n+ reviewed_at: '2026-05-04'\n+ note: Avi correction is score 3; not here, not there.\n@@\n - result_path: /home/avifenesh/projects/aviary/data/runs/session_summarize_for_avi/20260504-memory-gym-variety/qwen36-27b-autoround-64k/round-16-session-pairs-signal-sample/result.md\n- status: pending_avi_judge_review\n- note: Recent judge-draft batch; keep waiting for Avi unless verbal correction already exists.\n+ status: reviewed_excluded\n+ reviewed_at: '2026-05-04'\n+ note: Avi correction is exclude if leaked/contaminated; do not train on this example.\n*** End Patch\n",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "apply_patch: *** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/data/evals/judge-review-status.yml\n@@\n - result_path: /home/avifenesh/projects/aviary/data/runs/ses...",
"tool_name": "apply_patch"
},
"expected_probability": 0.5248817801475525,
"expected_source_event_id": "t9",
"field": "attempted_actions",
"rank": 20,
"reciprocal_rank": 0.05,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w53",
"top_candidates": [
{
"event": {
"command": "python3 scripts/judge-review-queue.py --write data/evals/judge-review-queues/current-pending.txt",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 scripts/judge-review-queue.py --write data/evals/judge-review-queues/current-pending.txt",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t12",
"support_probability": 0.9999039173126221
},
{
"event": {
"command": "python3 -m py_compile scripts/judge-review-queue.py scripts/review-judge-drafts.py scripts/build-scorecard-sft-dataset.py && cargo test eval_score --quiet",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 -m py_compile scripts/judge-review-queue.py scripts/review-judge-drafts.py scripts/build-scorecard-sft-dataset.py && cargo test eval_score --quiet",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t14",
"support_probability": 0.9998910427093506
},
{
"event": {
"command": "nl -ba data/evals/session_summarize_for_avi/scorecards.yml | sed -n '1020,1235p'",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "nl -ba data/evals/session_summarize_for_avi/scorecards.yml | sed -n '1020,1235p'",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t3",
"support_probability": 0.9996540546417236
},
{
"event": {
"command": "nl -ba data/evals/session_summarize_for_avi/scorecards.yml | sed -n '1230,1265p' && nl -ba data/evals/judge-review-status.yml | sed -n '70,140p'",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "nl -ba data/evals/session_summarize_for_avi/scorecards.yml | sed -n '1230,1265p' && nl -ba data/evals/judge-review-status.yml | sed -n '70,140p'",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t5",
"support_probability": 0.998793363571167
},
{
"event": {
"command": "python3 scripts/build-scorecard-sft-dataset.py --lane session_summarize_for_avi --out data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl && python3 scripts/build-scorecard-sft-dataset.py --lane session_summarize_for_avi --out data/training/session_summarize_for_avi/scorecard_sft_with_drafts.jsonl --include-drafts",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 scripts/build-scorecard-sft-dataset.py --lane session_summarize_for_avi --out data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl && python3 scripts...",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t13",
"support_probability": 0.9893474578857422
}
]
},
{
"expected_event": {
"command": "*** Begin Patch\n*** Add File: /home/avifenesh/projects/episodic-ingestion-compiler/data/examples/seed.jsonl\n+{\"id\":\"seed-0001-snake-tdd-validation-gap\",\"input\":{\"source_kind\":\"agent_trace\",\"source_uri\":\"pi://local/session/snake-tdd\",\"raw\":{\"source_events\":[{\"id\":\"u1\",\"role\":\"user\",\"timestamp\":\"2026-05-02T15:40:00Z\",\"text\":\"build a snake game that run on my machine locally without using browser, write the tests upfront\"},{\"id\":\"a1\",\"role\":\"assistant\",\"timestamp\":\"2026-05-02T15:41:00Z\",\"text\":\"I will write tests first, then implement the Snake logic and run cargo test.\"},{\"id\":\"t1\",\"role\":\"tool_call\",\"timestamp\":\"2026-05-02T15:42:00Z\",\"tool_id\":\"write_file\",\"file_paths\":[\"snake_game/test_snake.py\"],\"text\":\"create test suite for movement, food, wall collision, self collision\"},{\"id\":\"t2\",\"role\":\"tool_call\",\"timestamp\":\"2026-05-02T15:43:00Z\",\"tool_id\":\"run_command\",\"command_id\":\"cmd-cargo-test-1\",\"text\":\"cargo test\"},{\"id\":\"r2\",\"role\":\"tool_result\",\"timestamp\":\"2026-05-02T15:43:20Z\",\"tool_id\":\"run_command\",\"command_id\":\"cmd-cargo-test-1\",\"text\":\"tests passed\"},{\"id\":\"a2\",\"role\":\"assistant\",\"timestamp\":\"2026-05-02T15:44:00Z\",\"text\":\"The implementation is complete and tests pass.\"},{\"id\":\"u2\",\"role\":\"user\",\"timestamp\":\"2026-05-02T15:45:00Z\",\"text\":\"did you check also that the game work or just the tests?\"},{\"id\":\"r3\",\"role\":\"tool_result\",\"timestamp\":\"2026-05-02T15:46:00Z\",\"tool_id\":\"run_command\",\"command_id\":\"cmd-run-game-1\",\"text\":\"runtime smoke failed: game window starts but input does not begin movement\"}]}},\"expected\":{\"episode_hint\":\"snake-local-game-tdd\",\"objective_hint\":\"Build a local non-browser Snake game with tests written before implementation and verify both tests and playable runtime behavior.\",\"events\":[{\"event_id\":\"e1\",\"event_type\":\"user_request\",\"summary\":\"User requested a local non-browser Snake game and explicitly required tests upfront.\",\"provenance\":{\"source_event_ids\":[\"u1\"],\"timestamps\":[\"2026-05-02T15:40:00Z\"]},\"confidence\":0.98},{\"event_id\":\"e2\",\"event_type\":\"plan\",\"summary\":\"Assistant planned to write tests first, implement logic, and run cargo test.\",\"provenance\":{\"source_event_ids\":[\"a1\"],\"timestamps\":[\"2026-05-02T15:41:00Z\"]},\"confidence\":0.9},{\"event_id\":\"e3\",\"event_type\":\"file_change\",\"summary\":\"A test suite was created for movement, food, wall collision, and self collision.\",\"provenance\":{\"source_event_ids\":[\"t1\"],\"timestamps\":[\"2026-05-02T15:42:00Z\"],\"tool_ids\":[\"write_file\"],\"file_paths\":[\"snake_game/test_snake.py\"]},\"confidence\":0.92},{\"event_id\":\"e4\",\"event_type\":\"validation\",\"summary\":\"The unit test command passed.\",\"provenance\":{\"source_event_ids\":[\"t2\",\"r2\"],\"timestamps\":[\"2026-05-02T15:43:00Z\",\"2026-05-02T15:43:20Z\"],\"tool_ids\":[\"run_command\"],\"command_ids\":[\"cmd-cargo-test-1\"]},\"confidence\":0.95},{\"event_id\":\"e5\",\"event_type\":\"blocker\",\"summary\":\"A later runtime smoke showed the game did not actually start movement correctly despite passing tests.\",\"provenance\":{\"source_event_ids\":[\"u2\",\"r3\"],\"timestamps\":[\"2026-05-02T15:45:00Z\",\"2026-05-02T15:46:00Z\"],\"tool_ids\":[\"run_command\"],\"command_ids\":[\"cmd-run-game-1\"]},\"confidence\":0.88}],\"candidate_claims\":[{\"claim_id\":\"c1\",\"kind\":\"decided\",\"text\":\"The task acceptance requires runtime/playability validation, not only unit tests.\",\"support_event_ids\":[\"e1\",\"e5\"],\"evidence\":[{\"source_event_id\":\"u1\",\"quote\":\"build a snake game that run on my machine locally without using browser\"},{\"source_event_id\":\"u2\",\"quote\":\"did you check also that the game work or just the tests?\"}],\"truth_scope\":\"session_context\",\"promotion_hint\":\"strong_candidate\",\"confidence\":0.94},{\"claim_id\":\"c2\",\"kind\":\"validation_result\",\"text\":\"Unit tests passed, but the first runtime smoke failed to confirm playable behavior.\",\"support_event_ids\":[\"e4\",\"e5\"],\"evidence\":[{\"source_event_id\":\"r2\",\"quote\":\"tests passed\"},{\"source_event_id\":\"r3\",\"quote\":\"runtime smoke failed\"}],\"truth_scope\":\"historical\",\"promotion_hint\":\"candidate\",\"confidence\":0.9},{\"claim_id\":\"c3\",\"kind\":\"failed_because\",\"text\":\"The implementation was prematurely treated as complete after tests passed without sufficient runtime smoke validation.\",\"support_event_ids\":[\"e4\",\"e5\"],\"evidence\":[{\"source_event_id\":\"a2\",\"quote\":\"The implementation is complete and tests pass.\"},{\"source_event_id\":\"u2\",\"quote\":\"did you check also that the game work or just the tests?\"}],\"truth_scope\":\"session_context\",\"promotion_hint\":\"candidate\",\"confidence\":0.82}],\"decisions\":[{\"decision_id\":\"d1\",\"text\":\"Future similar tasks should track a runtime smoke check as a separate acceptance gate from unit tests.\",\"support_event_ids\":[\"e1\",\"e5\"],\"confidence\":0.86}],\"attempts\":[{\"attempt_id\":\"a1\",\"action\":\"Create tests before implementing Snake behavior.\",\"support_event_ids\":[\"e2\",\"e3\"],\"status\":\"succeeded\"},{\"attempt_id\":\"a2\",\"action\":\"Verify implementation with unit tests.\",\"support_event_ids\":[\"e4\"],\"status\":\"succeeded\"},{\"attempt_id\":\"a3\",\"action\":\"Verify playable runtime behavior.\",\"support_event_ids\":[\"e5\"],\"status\":\"failed\"}],\"outcomes\":[{\"outcome_id\":\"o1\",\"text\":\"Unit tests passed.\",\"support_event_ids\":[\"e4\"],\"status\":\"success\"},{\"outcome_id\":\"o2\",\"text\":\"Runtime behavior was not proven by the first test pass and required further fixing.\",\"support_event_ids\":[\"e5\"],\"status\":\"failure\"}],\"open_questions\":[{\"question_id\":\"q1\",\"text\":\"Which runtime input/rendering path prevented the game from starting movement?\",\"support_event_ids\":[\"e5\"]}],\"next_actions\":[{\"action_id\":\"n1\",\"text\":\"Add or run a local runtime smoke that verifies movement starts and input is handled.\",\"support_event_ids\":[\"e5\"],\"owner_hint\":\"agent\"}],\"warnings\":[{\"code\":\"source_gap\",\"message\":\"The trace does not include the final fixed code or successful runtime smoke, so do not promote final game correctness.\",\"source_event_ids\":[\"r3\"]}]}}\n+{\"id\":\"seed-0002-agnix-tool-release-triage\",\"input\":{\"source_kind\":\"repo_tool_trace\",\"source_uri\":\"github://agent-sh/agnix/issues/868\",\"raw\":{\"source_events\":[{\"id\":\"u1\",\"role\":\"user\",\"timestamp\":\"2026-05-05T21:47:55Z\",\"text\":\"check 868\"},{\"id\":\"t1\",\"role\":\"tool_call\",\"timestamp\":\"2026-05-05T21:47:58Z\",\"tool_id\":\"github\",\"text\":\"format issue 868 in agent-sh/agnix\"},{\"id\":\"r1\",\"role\":\"tool_result\",\"timestamp\":\"2026-05-05T21:48:00Z\",\"tool_id\":\"github\",\"text\":\"Issue #868: amp (Sourcegraph) gpt-5.5. Action required: review release notes for agnix validation rules, update config if validated fields changed, update research tracking, run workflow with update_baselines.\"},{\"id\":\"a1\",\"role\":\"assistant\",\"timestamp\":\"2026-05-05T21:48:08Z\",\"text\":\"Summary: GPT-5.5 is a model update for Amp deep mode; probably no agnix schema change unless config surfaces changed.\"},{\"id\":\"u2\",\"role\":\"user\",\"timestamp\":\"2026-05-05T21:50:00Z\",\"text\":\"so you get -1 on instruction follwing and he get +1\"},{\"id\":\"t2\",\"role\":\"tool_call\",\"timestamp\":\"2026-05-05T22:01:30Z\",\"tool_id\":\"edit\",\"file_paths\":[\".github/tool-release-baselines.json\"],\"text\":\"add changes_of_interest for amp config surfaces and relevant/irrelevant changes\"}]}},\"expected\":{\"episode_hint\":\"agnix-amp-release-triage\",\"objective_hint\":\"Triage auto-opened agnix issue #868 and identify whether Amp GPT-5.5 changes affect validation rules or tracking metadata.\",\"events\":[{\"event_id\":\"e1\",\"event_type\":\"user_request\",\"summary\":\"User asked to inspect issue #868.\",\"provenance\":{\"source_event_ids\":[\"u1\"],\"timestamps\":[\"2026-05-05T21:47:55Z\"]},\"confidence\":0.98},{\"event_id\":\"e2\",\"event_type\":\"tool_call\",\"summary\":\"GitHub issue #868 was fetched for agent-sh/agnix.\",\"provenance\":{\"source_event_ids\":[\"t1\",\"r1\"],\"timestamps\":[\"2026-05-05T21:47:58Z\",\"2026-05-05T21:48:00Z\"],\"tool_ids\":[\"github\"]},\"confidence\":0.96},{\"event_id\":\"e3\",\"event_type\":\"observation\",\"summary\":\"The issue describes an Amp model update and lists required triage actions around agnix validation rules and baselines.\",\"provenance\":{\"source_event_ids\":[\"r1\"],\"timestamps\":[\"2026-05-05T21:48:00Z\"],\"tool_ids\":[\"github\"]},\"confidence\":0.95},{\"event_id\":\"e4\",\"event_type\":\"observation\",\"summary\":\"Assistant summarized the issue as likely not requiring schema changes unless config surfaces changed.\",\"provenance\":{\"source_event_ids\":[\"a1\"],\"timestamps\":[\"2026-05-05T21:48:08Z\"]},\"confidence\":0.82},{\"event_id\":\"e5\",\"event_type\":\"file_change\",\"summary\":\"The baseline config was edited to add Amp changes_of_interest metadata.\",\"provenance\":{\"source_event_ids\":[\"t2\"],\"timestamps\":[\"2026-05-05T22:01:30Z\"],\"tool_ids\":[\"edit\"],\"file_paths\":[\".github/tool-release-baselines.json\"]},\"confidence\":0.93},{\"event_id\":\"e6\",\"event_type\":\"observation\",\"summary\":\"User noted an instruction-following penalty for the orchestrating assistant, while crediting another agent for following the intended fetch scope.\",\"provenance\":{\"source_event_ids\":[\"u2\"],\"timestamps\":[\"2026-05-05T21:50:00Z\"]},\"confidence\":0.86}],\"candidate_claims\":[{\"claim_id\":\"c1\",\"kind\":\"current_state\",\"text\":\"Issue #868 concerned Amp changing its tracked model baseline to GPT-5.5, not necessarily an agnix validation schema change.\",\"support_event_ids\":[\"e3\",\"e4\"],\"evidence\":[{\"source_event_id\":\"r1\",\"quote\":\"Issue #868: amp (Sourcegraph) gpt-5.5\"},{\"source_event_id\":\"a1\",\"quote\":\"probably no agnix schema change unless config surfaces changed\"}],\"truth_scope\":\"session_context\",\"promotion_hint\":\"candidate\",\"confidence\":0.84},{\"claim_id\":\"c2\",\"kind\":\"attempted\",\"text\":\"The agent updated .github/tool-release-baselines.json with changes_of_interest metadata for Amp.\",\"support_event_ids\":[\"e5\"],\"evidence\":[{\"source_event_id\":\"t2\",\"quote\":\"add changes_of_interest for amp config surfaces\"}],\"truth_scope\":\"historical\",\"promotion_hint\":\"strong_candidate\",\"confidence\":0.93},{\"claim_id\":\"c3\",\"kind\":\"has_next_action\",\"text\":\"The deterministic system should still verify whether the release baseline workflow and issue closure happened; the trace only shows the config edit.\",\"support_event_ids\":[\"e3\",\"e5\"],\"evidence\":[{\"source_event_id\":\"r1\",\"quote\":\"run workflow with update_baselines\"},{\"source_event_id\":\"t2\",\"quote\":\"add changes_of_interest\"}],\"truth_scope\":\"session_context\",\"promotion_hint\":\"candidate\",\"confidence\":0.8}],\"decisions\":[{\"decision_id\":\"d1\",\"text\":\"Treat model-version-only Amp releases as low schema risk unless config surfaces or validation inputs changed.\",\"support_event_ids\":[\"e3\",\"e4\",\"e5\"],\"confidence\":0.82}],\"attempts\":[{\"attempt_id\":\"a1\",\"action\":\"Fetch and inspect GitHub issue #868.\",\"support_event_ids\":[\"e1\",\"e2\"],\"status\":\"succeeded\"},{\"attempt_id\":\"a2\",\"action\":\"Add triage metadata for relevant Amp change types.\",\"support_event_ids\":[\"e5\"],\"status\":\"succeeded\"}],\"outcomes\":[{\"outcome_id\":\"o1\",\"text\":\"The trace contains a config edit but not final workflow/PR status.\",\"support_event_ids\":[\"e5\"],\"status\":\"partial\"}],\"open_questions\":[{\"question_id\":\"q1\",\"text\":\"Was the update_baselines workflow run and was issue #868 closed or linked to a PR?\",\"support_event_ids\":[\"e3\"]}],\"next_actions\":[{\"action_id\":\"n1\",\"text\":\"Check the PR or workflow result before promoting the triage as complete.\",\"support_event_ids\":[\"e3\",\"e5\"],\"owner_hint\":\"agent\"}],\"warnings\":[{\"code\":\"source_gap\",\"message\":\"The trace does not include the final CI/workflow result or issue closure state.\",\"source_event_ids\":[\"r1\",\"t2\"]}]}}\n+{\"id\":\"seed-0003-qwen-training-fastpath-decision\",\"input\":{\"source_kind\":\"training_trace\",\"source_uri\":\"file://aviary/qwen35-fastpath-experiment\",\"raw\":{\"source_events\":[{\"id\":\"u1\",\"role\":\"user\",\"timestamp\":\"2026-05-06T23:00:00Z\",\"text\":\"we trained 27b on this machine, how 9b oom on 2048?\"},{\"id\":\"a1\",\"role\":\"assistant\",\"timestamp\":\"2026-05-06T23:02:00Z\",\"text\":\"The 9B run is using a different path than the 27B smoke: full modules, different activation memory, and possibly non-fast fallback.\"},{\"id\":\"t1\",\"role\":\"tool_call\",\"timestamp\":\"2026-05-06T23:10:00Z\",\"tool_id\":\"pip\",\"command_id\":\"cmd-install-fla\",\"text\":\"install flash-linear-attention, fla-core, causal-conv1d with CUDA 13 headers workaround\"},{\"id\":\"r1\",\"role\":\"tool_result\",\"timestamp\":\"2026-05-06T23:20:00Z\",\"tool_id\":\"python\",\"command_id\":\"cmd-check-fastpath\",\"text\":\"flash-linear-attention==0.5.0, fla-core==0.5.0, causal-conv1d==1.6.1, is_fast_path_available True\"},{\"id\":\"t2\",\"role\":\"tool_call\",\"timestamp\":\"2026-05-06T23:30:00Z\",\"tool_id\":\"train\",\"command_id\":\"cmd-train-v06\",\"text\":\"train qwen35-9b context ops v0.6 invariants fla full 1792 60step\"},{\"id\":\"r2\",\"role\":\"tool_result\",\"timestamp\":\"2026-05-06T23:38:00Z\",\"tool_id\":\"train\",\"command_id\":\"cmd-train-v06\",\"text\":\"train_loss 0.2719536894311508 runtime 408.7794s peak_allocated 22.1617 GiB peak_reserved 22.2383 GiB\"},{\"id\":\"r3\",\"role\":\"tool_result\",\"timestamp\":\"2026-05-06T23:45:00Z\",\"tool_id\":\"eval\",\"command_id\":\"cmd-eval-v06\",\"text\":\"challenge raw-domain-valid 10/11 repaired-valid 11/11; remaining raw failure emitted resolution ask instead of ask_user\"}]}},\"expected\":{\"episode_hint\":\"qwen35-9b-fla-training-setup\",\"objective_hint\":\"Diagnose why a Qwen3.5 9B LoRA run OOMed and establish a stable fast-path training setup for episodic context operations.\",\"events\":[{\"event_id\":\"e1\",\"event_type\":\"user_request\",\"summary\":\"User questioned why the smaller 9B model OOMed when a 27B smoke had trained locally.\",\"provenance\":{\"source_event_ids\":[\"u1\"],\"timestamps\":[\"2026-05-06T23:00:00Z\"]},\"confidence\":0.98},{\"event_id\":\"e2\",\"event_type\":\"observation\",\"summary\":\"Assistant identified that memory behavior differed because the 9B run used different module coverage, activation path, and possibly non-fast fallback.\",\"provenance\":{\"source_event_ids\":[\"a1\"],\"timestamps\":[\"2026-05-06T23:02:00Z\"]},\"confidence\":0.82},{\"event_id\":\"e3\",\"event_type\":\"attempt\",\"summary\":\"The fast-path packages were installed using a CUDA 13 headers workaround.\",\"provenance\":{\"source_event_ids\":[\"t1\"],\"timestamps\":[\"2026-05-06T23:10:00Z\"],\"tool_ids\":[\"pip\"],\"command_ids\":[\"cmd-install-fla\"]},\"confidence\":0.9},{\"event_id\":\"e4\",\"event_type\":\"validation\",\"summary\":\"The Qwen3.5 fast path was verified as available with FLA 0.5.0 and causal-conv1d 1.6.1.\",\"provenance\":{\"source_event_ids\":[\"r1\"],\"timestamps\":[\"2026-05-06T23:20:00Z\"],\"tool_ids\":[\"python\"],\"command_ids\":[\"cmd-check-fastpath\"]},\"confidence\":0.98},{\"event_id\":\"e5\",\"event_type\":\"validation\",\"summary\":\"Qwen3.5 9B v0.6 trained for 60 steps at 1792 context with peak reserved memory around 22.24 GiB.\",\"provenance\":{\"source_event_ids\":[\"t2\",\"r2\"],\"timestamps\":[\"2026-05-06T23:30:00Z\",\"2026-05-06T23:38:00Z\"],\"tool_ids\":[\"train\"],\"command_ids\":[\"cmd-train-v06\"]},\"confidence\":0.97},{\"event_id\":\"e6\",\"event_type\":\"outcome\",\"summary\":\"Challenge eval reached 10/11 raw-domain-valid and 11/11 after deterministic repair, with one enum drift failure.\",\"provenance\":{\"source_event_ids\":[\"r3\"],\"timestamps\":[\"2026-05-06T23:45:00Z\"],\"tool_ids\":[\"eval\"],\"command_ids\":[\"cmd-eval-v06\"]},\"confidence\":0.97}],\"candidate_claims\":[{\"claim_id\":\"c1\",\"kind\":\"worked_because\",\"text\":\"The Qwen3.5 9B training path became viable after the FLA/causal-conv fast path was installed and verified.\",\"support_event_ids\":[\"e3\",\"e4\",\"e5\"],\"evidence\":[{\"source_event_id\":\"r1\",\"quote\":\"is_fast_path_available True\"},{\"source_event_id\":\"r2\",\"quote\":\"peak_reserved 22.2383 GiB\"}],\"truth_scope\":\"historical\",\"promotion_hint\":\"strong_candidate\",\"confidence\":0.92},{\"claim_id\":\"c2\",\"kind\":\"validation_result\",\"text\":\"The v0.6 adapter still had one raw enum drift issue: resolution ask instead of ask_user.\",\"support_event_ids\":[\"e6\"],\"evidence\":[{\"source_event_id\":\"r3\",\"quote\":\"remaining raw failure emitted resolution ask instead of ask_user\"}],\"truth_scope\":\"historical\",\"promotion_hint\":\"strong_candidate\",\"confidence\":0.97},{\"claim_id\":\"c3\",\"kind\":\"has_next_action\",\"text\":\"The next training material should include enum/invariant correction examples rather than more generic rows.\",\"support_event_ids\":[\"e6\"],\"evidence\":[{\"source_event_id\":\"r3\",\"quote\":\"remaining raw failure emitted resolution ask instead of ask_user\"}],\"truth_scope\":\"session_context\",\"promotion_hint\":\"candidate\",\"confidence\":0.86}],\"decisions\":[{\"decision_id\":\"d1\",\"text\":\"Use the verified FLA full path as the stable Qwen3.5 9B training setup, while keeping deterministic repair in the eval loop.\",\"support_event_ids\":[\"e4\",\"e5\",\"e6\"],\"confidence\":0.91}],\"attempts\":[{\"attempt_id\":\"a1\",\"action\":\"Install and verify Qwen3.5 fast-path dependencies.\",\"support_event_ids\":[\"e3\",\"e4\"],\"status\":\"succeeded\"},{\"attempt_id\":\"a2\",\"action\":\"Train Qwen3.5 9B context-ops v0.6 for 60 steps.\",\"support_event_ids\":[\"e5\"],\"status\":\"succeeded\"},{\"attempt_id\":\"a3\",\"action\":\"Evaluate v0.6 on the challenge split.\",\"support_event_ids\":[\"e6\"],\"status\":\"succeeded\"}],\"outcomes\":[{\"outcome_id\":\"o1\",\"text\":\"Training fit within 24 GiB class VRAM with about 22.24 GiB peak reserved.\",\"support_event_ids\":[\"e5\"],\"status\":\"success\"},{\"outcome_id\":\"o2\",\"text\":\"Eval was valid after repair but not perfectly raw-valid due to enum drift.\",\"support_event_ids\":[\"e6\"],\"status\":\"partial\"}],\"open_questions\":[{\"question_id\":\"q1\",\"text\":\"Does the same fast path stay stable for larger context or broader adapter coverage?\",\"support_event_ids\":[\"e5\"]}],\"next_actions\":[{\"action_id\":\"n1\",\"text\":\"Add correction rows for enum/invariant failures and train a short v0.7 comparison from the base model.\",\"support_event_ids\":[\"e6\"],\"owner_hint\":\"agent\"}],\"warnings\":[{\"code\":\"policy_owned_by_deterministic_system\",\"message\":\"The model can propose normalized training-state claims, but the verifier should own final acceptance of adapter quality.\",\"source_event_ids\":[\"r3\"]}]}}\n*** End Patch\n",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "apply_patch: *** Begin Patch\n*** Add File: /home/avifenesh/projects/episodic-ingestion-compiler/data/examples/seed.jsonl\n+{\"id\":\"seed-0001-snake-tdd-validation-gap\",\"input\":{\"so...",
"tool_name": "apply_patch"
},
"expected_probability": 0.4651445150375366,
"expected_source_event_id": "t19",
"field": "attempted_actions",
"rank": 20,
"reciprocal_rank": 0.05,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w88",
"top_candidates": [
{
"event": {
"command": "test -e /home/avifenesh/projects/episodic-ingestion-compiler; echo $?",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "test -e /home/avifenesh/projects/episodic-ingestion-compiler; echo $?",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t6",
"support_probability": 0.9998511075973511
},
{
"event": {
"command": "mkdir -p /home/avifenesh/projects/episodic-ingestion-compiler/{src/episodic_ingestion_compiler,scripts,schemas,data/examples,data/sft,docs,tests}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "mkdir -p /home/avifenesh/projects/episodic-ingestion-compiler/{src/episodic_ingestion_compiler,scripts,schemas,data/examples,data/sft,docs,tests}",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t12",
"support_probability": 0.9995827078819275
},
{
"event": {
"command": "python3 - <<'PY'\ntry:\n import jsonschema\n print('jsonschema ok')\nexcept Exception as e:\n print(f'jsonschema missing: {e}')\nPY",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 - <<'PY'\ntry:\n import jsonschema\n print('jsonschema ok')\nexcept Exception as e:\n print(f'jsonschema missing: {e}')\nPY",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t13",
"support_probability": 0.9994966983795166
},
{
"event": {
"command": "{\"plan\": [{\"step\": \"Create clean repo scaffold\", \"status\": \"in_progress\"}, {\"step\": \"Define ingestion schema and contract\", \"status\": \"pending\"}, {\"step\": \"Add seed training examples and validators\", \"status\": \"pending\"}, {\"step\": \"Run validation and commit\", \"status\": \"pending\"}, {\"step\": \"Report next training path\", \"status\": \"pending\"}]}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "{\"plan\": [{\"step\": \"Create clean repo scaffold\", \"status\": \"in_progress\"}, {\"step\": \"Define ingestion schema and contract\", \"status\": \"pending\"}, {\"step\": \"Add seed training exa...",
"tool_name": "update_plan"
},
"label": 1,
"source_event_id": "t3",
"support_probability": 0.99935382604599
},
{
"event": {
"command": "git status --short",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "git status --short",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t7",
"support_probability": 0.99935382604599
}
]
},
{
"expected_event": {
"command": "python3 << 'EOF'\nimport json, re, sys\n\nwith open('/tmp/valkey_prs_batch4.json') as f:\n raw = f.read()\n\n# Split by PR markers\nprs = re.split(r'=== PR #(\\d+) ===\\n', raw)[1:]\n# pairs: [num, content, num, content, ...]\n\nfor i in range(0, len(prs), 2):\n num = prs[i]\n content = prs[i+1]\n # Split content at --- inline comments ---\n parts = content.split('--- inline comments ---\\n', 1)\n pr_json_text = parts[0].strip()\n inline_text = parts[1].strip() if len(parts) > 1 else ''\n \n try:\n pr = json.loads(pr_json_text)\n except Exception as e:\n print(f\"PR {num} json err: {e}\")\n continue\n \n print(f\"\\n===== PR #{num} =====\")\n print(f\"Title: {pr.get('title')}\")\n print(f\"Base: {pr.get('baseRefName')}\")\n print(f\"Author: {pr.get('author', {}).g",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 << 'EOF'\nimport json, re, sys\n\nwith open('/tmp/valkey_prs_batch4.json') as f:\n raw = f.read()\n\n# Split by PR markers\nprs = re.split(r'=== PR #(\\d+) ===\\n', raw)[1:]\n#...",
"tool_name": "Bash"
},
"expected_probability": 0.3988746404647827,
"expected_source_event_id": "t7",
"field": "attempted_actions",
"rank": 20,
"reciprocal_rank": 0.05,
"row_id": "pta-claude_code-agent-a55fe1edceb764d9e::w0",
"top_candidates": [
{
"event": {
"command": "cd /home/avifenesh/projects/valkey && git log --all --grep=\"#3469\" --oneline | head -5 && echo \"---\" && git log --all --oneline | grep -i \"3469\" | head -5 && echo \"---forkless branch---\" && git branch -a | grep -i forkless | head",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cd /home/avifenesh/projects/valkey && git log --all --grep=\"#3469\" --oneline | head -5 && echo \"---\" && git log --all --oneline | grep -i \"3469\" | head -5 && echo \"---forkless b...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t14",
"support_probability": 0.999595582485199
},
{
"event": {
"command": "cd /home/avifenesh/projects/valkey && git fetch origin +refs/pull/3469/head:refs/remotes/origin/pr-3469 2>&1 | tail -3 && git log --format='%H %s%n%b' -1 origin/pr-3469 2>&1 | head -40",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cd /home/avifenesh/projects/valkey && git fetch origin +refs/pull/3469/head:refs/remotes/origin/pr-3469 2>&1 | tail -3 && git log --format='%H %s%n%b' -1 origin/pr-3469 2>&1 | h...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t16",
"support_probability": 0.999555766582489
},
{
"event": {
"command": "ls -la /tmp/valkey_prs_batch4.json && head -100 /tmp/valkey_prs_batch4.json",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "ls -la /tmp/valkey_prs_batch4.json && head -100 /tmp/valkey_prs_batch4.json",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t5",
"support_probability": 0.9994115829467773
},
{
"event": {
"command": "Read /home/avifenesh/.claude/projects/-home-avifenesh-projects-valkey-skills/ff8a0875-8828-483a-a804-8fd20d09aecb/tool-results/bq6pp40yy.txt",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "Read /home/avifenesh/.claude/projects/-home-avifenesh-projects-valkey-skills/ff8a0875-8828-483a-a804-8fd20d09aecb/tool-results/bq6pp40yy.txt",
"tool_name": "Read"
},
"label": 1,
"source_event_id": "t9",
"support_probability": 0.9987157583236694
},
{
"event": {
"command": "for pr in 3498 3495 3472 3471 3469 3464 3463 3462 3461 3460; do\n echo \"=== PR #$pr ===\"\n gh pr view $pr --repo valkey-io/valkey --json title,body,reviews,comments,labels,baseRefName,mergeCommit,author 2>&1\n echo \"--- inline comments ---\"\n gh api repos/valkey-io/valkey/pulls/$pr/comments --paginate 2>&1\n echo\ndone > /tmp/valkey_prs_batch4.json 2>&1\nwc -l /tmp/valkey_prs_batch4.json",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "for pr in 3498 3495 3472 3471 3469 3464 3463 3462 3461 3460; do\n echo \"=== PR #$pr ===\"\n gh pr view $pr --repo valkey-io/valkey --json title,body,reviews,comments,labels,baseR...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t3",
"support_probability": 0.9980732202529907
}
]
},
{
"expected_event": {
"command": "for sha in 5157255b82002a79fd212de9d9ba0fe08812849f 6b384e9a5b6cd4e4f8198f5df22a74feb1b70e0f 6c329dfe2c02d09b276fd5617acdb31e6169ebc1 8caa29298c51b4bb498418515801dd0720ebce62 5deebd0f131826124b3ab6231fafb5f9b188523b 6b85ca4fffff51de5f8cfcc6ce239900a48924d0 1e2ca5ec5dd6f97215e7acbcee14e89ea56a9e02 6414720504f689a3fc699338bf22565bb70c3829 9f0106755b177f8eea3d62eed6f28f20fb54d459 a3a839972ff10a68b1a7069d70d8ce1fffb8f02a; do echo \"=== $sha ===\"; git -C /home/avifenesh/projects/valkey show --format='%s%n%b%n---END---' -s $sha 2>&1 | head -30; done",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "for sha in 5157255b82002a79fd212de9d9ba0fe08812849f 6b384e9a5b6cd4e4f8198f5df22a74feb1b70e0f 6c329dfe2c02d09b276fd5617acdb31e6169ebc1 8caa29298c51b4bb498418515801dd0720ebce62 5d...",
"tool_name": "Bash"
},
"expected_probability": 0.709019124507904,
"expected_source_event_id": "t12",
"field": "attempted_actions",
"rank": 20,
"reciprocal_rank": 0.05,
"row_id": "pta-claude_code-agent-a5a805ef49f3ca63c::w0",
"top_candidates": [
{
"event": {
"command": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do gh api repos/valkey-io/valkey/pulls/$pr/comments --paginate > inline-$pr.json 2>&1 & done; wait; ls -la inline-*",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do gh api repos/valkey-io/valkey/pulls/$pr/comments --paginate > inline-$pr.json 2>&1 & done; w...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t4",
"support_probability": 0.9994471669197083
},
{
"event": {
"command": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do echo \"=== PR $pr inline-comments ===\"; jq -r '.[] | \"@\\(.user.login) on \\(.path):\\(.line // .original_line): \\((.body // \"\")[:250])\"' inline-$pr.json 2>/dev/null | head -30; done 2>&1 | head -300",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do echo \"=== PR $pr inline-comments ===\"; jq -r '.[] | \"@\\(.user.login) on \\(.path):\\(.line //...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t11",
"support_probability": 0.9993928670883179
},
{
"event": {
"command": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do echo \"=== PR $pr reviews+comments ===\"; jq -r '.reviews[] | \"[review] [\\(.state)] @\\(.author.login): \\((.body // \"\")[:300])\"' pr-$pr.json; jq -r '.comments[] | \"[comment] @\\(.author.login): \\((.body // \"\")[:300])\"' pr-$pr.json; done 2>&1 | head -500",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do echo \"=== PR $pr reviews+comments ===\"; jq -r '.reviews[] | \"[review] [\\(.state)] @\\(.author...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t9",
"support_probability": 0.9992678761482239
},
{
"event": {
"command": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do echo \"=== PR $pr ===\"; jq -r '{title, author: .author.login, baseRefName, labels: [.labels[].name], mergeCommit: .mergeCommit.oid, body: (.body|.[:1500])}' pr-$pr.json; done 2>&1 | head -400",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cd /tmp/pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do echo \"=== PR $pr ===\"; jq -r '{title, author: .author.login, baseRefName, labels: [.labels[]...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t7",
"support_probability": 0.9988664388656616
},
{
"event": {
"command": "cd /tmp && mkdir -p pr-survey && cd pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do gh pr view $pr --repo valkey-io/valkey --json title,body,reviews,comments,labels,baseRefName,mergeCommit,author > pr-$pr.json 2>&1 & done; wait; ls -la",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cd /tmp && mkdir -p pr-survey && cd pr-survey && for pr in 3343 3342 3340 3338 3336 3333 3331 3329 3325 3324; do gh pr view $pr --repo valkey-io/valkey --json title,body,reviews...",
"tool_name": "Bash"
},
"label": 1,
"source_event_id": "t3",
"support_probability": 0.9987550973892212
}
]
},
{
"expected_event": {
"command": "ROOT=/home/avifenesh/projects/aviary; RUN_LOG=\"$ROOT/data/runs/qwen35-moe-8k-triton-$(date +%Y%m%d-%H%M%S).log\"; export PATH=\"$ROOT/.venv/bin:$PATH\" PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0; echo \"log=$RUN_LOG\"; sudo -n systemd-run --scope -p MemoryMax=32G -p MemorySwapMax=0 \"$ROOT/.venv/bin/vllm\" serve \"$ROOT/models/qwen3.6-35b-a3b-nvfp4\" --host 127.0.0.1 --port 8000 --served-model-name aviary-local-qwen --trust-remote-code --language-model-only --skip-mm-profiling --chat-template \"$ROOT/configs/qwen-codex-chat-template.jinja\" --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 8192 --kv-cache-dtype fp8 --enable-prefix-caching --gpu-memory-utilization 0.92 --max-num-seqs 1 --max-num-batched-to",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "ROOT=/home/avifenesh/projects/aviary; RUN_LOG=\"$ROOT/data/runs/qwen35-moe-8k-triton-$(date +%Y%m%d-%H%M%S).log\"; export PATH=\"$ROOT/.venv/bin:$PATH\" PYTORCH_CUDA_ALLOC_CONF=expa...",
"tool_name": "exec_command"
},
"expected_probability": 0.4012185037136078,
"expected_source_event_id": "t10",
"field": "attempted_actions",
"rank": 19,
"reciprocal_rank": 0.05263157894736842,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w14",
"top_candidates": [
{
"event": {
"command": "{\"session_id\": 92479, \"chars\": \"\\u0003\", \"yield_time_ms\": 1000, \"max_output_tokens\": 20000}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "{\"session_id\": 92479, \"chars\": \"\\u0003\", \"yield_time_ms\": 1000, \"max_output_tokens\": 20000}",
"tool_name": "write_stdin"
},
"label": 1,
"source_event_id": "t3",
"support_probability": 0.9996079802513123
},
{
"event": {
"command": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 30000}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 30000}",
"tool_name": "write_stdin"
},
"label": 1,
"source_event_id": "t12",
"support_probability": 0.9992678761482239
},
{
"event": {
"command": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 6000}",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "{\"session_id\": 1703, \"chars\": \"\", \"yield_time_ms\": 1000, \"max_output_tokens\": 6000}",
"tool_name": "write_stdin"
},
"label": 1,
"source_event_id": "t17",
"support_probability": 0.9992678761482239
},
{
"event": {
"command": "nvidia-smi --query-gpu=memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp sport = :8000",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "nvidia-smi --query-gpu=memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp sport = :8000",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t13",
"support_probability": 0.9985895752906799
},
{
"event": {
"command": "nvidia-smi --query-gpu=name,memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp 'sport = :8000' || true",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "nvidia-smi --query-gpu=name,memory.used,memory.free,utilization.gpu --format=csv,noheader,nounits && free -h && ss -ltnp 'sport = :8000' || true",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t18",
"support_probability": 0.998245358467102
}
]
},
{
"expected_event": {
"command": "*** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/scripts/build-scorecard-sft-dataset.py\n@@\n-By default this exports only canonical Avi scorecards. Use\n---include-corrected-drafts to add draft files that Avi corrected, and\n---include-drafts only for research/debug datasets.\n+By default this exports only canonical Avi scorecards. Use\n+--include-corrected-drafts to add draft files that Avi corrected, and\n+--include-drafts only for research/debug datasets. Individual cards can set\n+exclude_from_sft: true when they are useful as reviewed eval evidence but\n+unsafe or unhelpful as training targets.\n@@\n def target_yaml(card: dict[str, Any]) -> str:\n@@\n return yaml.safe_dump(target, sort_keys=False, allow_unicode=False).strip() + \"\\n\"\n+\n+\n+def is_excluded_from_sft(card: dict[str, Any]) -> bool:\n+ if bool(card.get(\"exclude_from_sft\")):\n+ return True\n+ eligibility = str(card.get(\"training_eligibility\", \"\")).strip().lower()\n+ return eligibility in {\"exclude\", \"excluded\", \"do_not_train\", \"not_for_training\"}\n@@\n records = source_records(args.lane, args.include_corrected_drafts, args.include_drafts)\n for source_file, card, source_kind in records:\n stats[f\"source:{source_kind}\"] += 1\n+ if is_excluded_from_sft(card):\n+ stats[f\"excluded_from_sft:{source_kind}\"] += 1\n+ skipped.append(\n+ {\n+ \"source_file\": str(source_file),\n+ \"run_id\": str(card.get(\"run_id\")),\n+ \"case_id\": str(card.get(\"case_id\")),\n+ \"reason\": \"card marked exclude_from_sft\",\n+ }\n+ )\n+ continue\n result_path = result_path_for_card(args.lane, card)\n*** End Patch\n",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "apply_patch: *** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/scripts/build-scorecard-sft-dataset.py\n@@\n-By default this exports only canonical Avi scorecards. U...",
"tool_name": "apply_patch"
},
"expected_probability": 0.6141626238822937,
"expected_source_event_id": "t3",
"field": "attempted_actions",
"rank": 19,
"reciprocal_rank": 0.05263157894736842,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w32",
"top_candidates": [
{
"event": {
"command": "wc -l data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl data/training/session_summarize_for_avi/scorecard_sft_with_drafts.jsonl",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "wc -l data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl data/training/session_summarize_for_avi/scorecard_sft_with_drafts.jsonl",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t13",
"support_probability": 0.9992903470993042
},
{
"event": {
"command": "python3 -m py_compile scripts/build-scorecard-sft-dataset.py && scripts/build-scorecard-sft-dataset.py --out data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl --include-corrected-drafts",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 -m py_compile scripts/build-scorecard-sft-dataset.py && scripts/build-scorecard-sft-dataset.py --out data/training/session_summarize_for_avi/scorecard_sft_avi_verified.j...",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t9",
"support_probability": 0.9989350438117981
},
{
"event": {
"command": "sed -n '1,120p' data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl.manifest.yml",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "sed -n '1,120p' data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl.manifest.yml",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t14",
"support_probability": 0.9987550973892212
},
{
"event": {
"command": "scripts/build-scorecard-sft-dataset.py --out data/training/session_summarize_for_avi/scorecard_sft_with_drafts.jsonl --include-drafts",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "scripts/build-scorecard-sft-dataset.py --out data/training/session_summarize_for_avi/scorecard_sft_with_drafts.jsonl --include-drafts",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t10",
"support_probability": 0.9986329674720764
},
{
"event": {
"command": "rg -n \"round-04-workbench|round-05-one-bash|round-06-qwen9\" data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl.manifest.yml",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "rg -n \"round-04-workbench|round-05-one-bash|round-06-qwen9\" data/training/session_summarize_for_avi/scorecard_sft_avi_verified.jsonl data/training/session_summarize_for_avi/scor...",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t15",
"support_probability": 0.9969245791435242
}
]
},
{
"expected_event": {
"command": "*** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml\n@@\n-review_status: needs_avi_review\n+review_status: avi_corrected_partial\n+reviewed_by: avi\n+review_note: >-\n+ Avi rated the main Aviary web_fetch result as 2-3. This is not a zero because\n+ it found the most-starred agent-sh repo, but it is still below usable because\n+ it missed the real open-work queue and over-presented incomplete research.\n@@\n - run_id: \"20260503-github-web-research/aviary-chat-once-webfetch\"\n model_label: qwen35-9b\n harness: aviary-chat-once\n case_id: round-01-github-open-issues-agent-sh-stars\n- decision: draft_negative_mixed_webfetch_result\n+ review_status: avi_corrected\n+ reviewed_by: avi\n+ decision: verified_low_mixed_webfetch_result_after_correction\n+ avi_score_range: \"2-3\"\n scores:\n- overall_usefulness: 2\n+ overall_usefulness: 2\n@@\n do_not_train_as_success: 1\n+ avi_feedback: >-\n+ I would give it 2-3.\n summary: >-\n Mixed but mostly negative web-research sample. The model used Aviary's\n web_search/web_fetch tools and correctly found `agent-sh/agentsys` as the\n most-starred repo with 777 stars. It failed the more important work-queue\n part: it did not inspect `avifenesh/*` properly, missed open\n `agent-sh/agnix` issues/PRs involving Avi, relied on shallow GitHub pages,\n- and presented incomplete findings too confidently.\n+ and presented incomplete findings too confidently. Avi corrected this to\n+ a 2-3 range; encode conservatively as 2 while preserving the range.\n*** End Patch\n",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "apply_patch: *** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml\n@@\n-review_s...",
"tool_name": "apply_patch"
},
"expected_probability": 0.7799928784370422,
"expected_source_event_id": "t7",
"field": "attempted_actions",
"rank": 19,
"reciprocal_rank": 0.05263157894736842,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w34",
"top_candidates": [
{
"event": {
"command": "git diff -- data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "git diff -- data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t10",
"support_probability": 0.999841570854187
},
{
"event": {
"command": "git diff --check",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "git diff --check",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t11",
"support_probability": 0.9997040629386902
},
{
"event": {
"command": "git add data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml && git commit -m \"Mark web research eval corrected\"",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "git add data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml && git commit -m \"Mark web research eval corrected\"",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t16",
"support_probability": 0.999555766582489
},
{
"event": {
"command": "rg -n \"review_status|aviary-chat-once|overall_usefulness|avi_feedback|decision|summary\" data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "rg -n \"review_status|aviary-chat-once|overall_usefulness|avi_feedback|decision|summary\" data/runs/web_research_for_avi/20260503-github-web-research/codex-scorecards-draft.yml",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t3",
"support_probability": 0.999480664730072
},
{
"event": {
"command": "git status --short",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "git status --short",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t18",
"support_probability": 0.9994115829467773
}
]
},
{
"expected_event": {
"command": "*** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/data/evals/personal_agent_for_avi/scorecards.yml\n@@\n - pr_workflow_reference\n - bookkeeping_recovery_reference\n - shell_quoting_negative_reference\n - safe_merge_policy_reference\n+- id: score-20260504T144215-qwen27-agnix-claude-code-triage\n+ created_at: 2026-05-04T14:42:15+03:00\n+ rater: avi\n+ lane: personal_agent_for_avi\n+ run_id: 20260504-live-agnix-claude-code-triage\n+ model_label: aviary-local-qwen\n+ case_id: agnix-claude-code-issue-850-triage\n+ result_path: /home/avifenesh/projects/aviary/data/runs/personal_agent_for_avi/20260504-live-agnix-claude-code-triage/aviary-local-qwen/agnix-claude-code-issue-850-triage/input.jsonl\n+ input_path: /home/avifenesh/projects/aviary/data/runs/personal_agent_for_avi/20260504-live-agnix-claude-code-triage/aviary-local-qwen/agnix-claude-code-issue-850-triage/input.jsonl\n+ decision: useful_success_with_branch_and_quoting_failures\n+ scores:\n+ agentic_behavior: 4\n+ bookkeeping: 2\n+ grounding_no_hallucination: 4\n+ instruction_following: 4\n+ judgment_prioritization: 3\n+ latency_acceptability: 4\n+ long_context_handling: 4\n+ output_quality: 3\n+ overall_usefulness: 4\n+ recovery: 4\n+ safety_restraint: 3\n+ terminal_usage: 3\n+ tone_intent_understanding: 4\n+ tool_use: 3\n+ notes: >-\n+ Avi initially judged this as another useful session: helpful, fast, good\n+ terminal/tool work, and good task understanding. After review, Avi corrected\n+ that the agent did make a mistake. The material mistake was branch hygiene:\n+ it opened the Claude Code PR while still based on the previous OpenCode PR\n+ branch state, so PR #863 initially included old OpenCode commits and\n+ triggered stale Copilot/Gemini review comments. The agent also repeated the\n+ shell quoting bug from the previous session: backticks in a double-quoted\n+ `gh pr create --body` command were interpreted by the shell, leaving a\n+ mangled PR body. The final result was still useful because it recovered after\n+ Avi pointed out the wrong branch, rebased onto main, produced a clean diff,\n+ waited for all checks, and merged PR #863 with issue #850 closed.\n+ evidence:\n+ - The agent correctly found 13 open issues, separated 5 auto issues from 8 manual issues, and selected #850 as the first auto issue to process.\n+ - It correctly classified Claude Code v2.1.126 as agnix-irrelevant, updated the baseline and Last Reviewed date, ran rule bookkeeping, opened PR #863, waited for CI/reviews, and eventually merged it.\n+ - The agent created the new branch before updating local main; `git log main..HEAD` showed the previous OpenCode commits in the Claude Code PR.\n+ - Avi corrected the branch mistake with \"because you need to move to main you are on a wrong branch\"; the agent switched to main, pulled, rebased, and reduced the PR to one clean Claude Code commit.\n+ - PR #863 body still shows shell quoting damage in the Changes bullets because Markdown backticks were interpreted by the shell during `gh pr create`.\n+ - All checks passed before merge, PR #863 merged, and issue #850 closed.\n+ failure_modes:\n+ - started_pr_from_wrong_base_branch\n+ - stale_prior_pr_commits_included_initially\n+ - stale_bot_review_noise_from_dirty_base\n+ - shell_quoting_backticks_mangled_pr_body\n+ - did_not_repair_mangled_pr_body\n+ - repeated_same_pr_body_quoting_failure\n+ - gh_pr_check_typo\n+ - attempted_review_dismissal_with_wrong_api\n+ tuning_uses:\n+ - preference_dataset\n+ - eval_regression\n+ - judge_training\n+ - supervised_finetune_example\n+ - positive_agentic_terminal_reference\n+ - branch_hygiene_negative_reference\n+ - pr_workflow_recovery_reference\n+ - shell_quoting_negative_reference\n+ - owner_correction_recovery_reference\n*** End Patch\n",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "apply_patch: *** Begin Patch\n*** Update File: /home/avifenesh/projects/aviary/data/evals/personal_agent_for_avi/scorecards.yml\n@@\n - pr_workflow_reference\n - bookkeeping_rec...",
"tool_name": "apply_patch"
},
"expected_probability": 0.48121002316474915,
"expected_source_event_id": "t10",
"field": "attempted_actions",
"rank": 19,
"reciprocal_rank": 0.05263157894736842,
"row_id": "pta-codex-rollout-2026-04-30T19-43-19-019ddf45-fdda-7fb2-8764-8162ba11dc24::w45",
"top_candidates": [
{
"event": {
"command": "cargo test eval_score --quiet",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "cargo test eval_score --quiet",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t13",
"support_probability": 0.9996849298477173
},
{
"event": {
"command": "tail -n 20 data/evals/personal_agent_for_avi/scorecards.yml",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "tail -n 20 data/evals/personal_agent_for_avi/scorecards.yml",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t5",
"support_probability": 0.9995827078819275
},
{
"event": {
"command": "date --iso-8601=seconds",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "date --iso-8601=seconds",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t4",
"support_probability": 0.999555766582489
},
{
"event": {
"command": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\np=Path('data/evals/personal_agent_for_avi/scorecards.yml')\ndata=yaml.safe_load(p.read_text())\nprint(data['lane'], len(data['scorecards']))\nfor c in data['scorecards']:\n print(c['id'], c['decision'], c['scores'].get('overall_usefulness'), c['scores'].get('bookkeeping'))\nPY",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\np=Path('data/evals/personal_agent_for_avi/scorecards.yml')\ndata=yaml.safe_load(p.read_text())\nprint(data['lane'], len(data[...",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t14",
"support_probability": 0.9985449314117432
},
{
"event": {
"command": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\nroot=Path('/home/avifenesh/projects/aviary')\ntotal=0\nfor p in sorted((root/'data/evals').glob('*/scorecards.yml')):\n data=yaml.safe_load(p.read_text()) or {}\n n=len(data.get('scorecards') or [])\n total+=n\n print(data.get('lane') or p.parent.name, n)\nprint('total', total)\nPY",
"is_context_tool_echo": false,
"role": "tool_call",
"success": null,
"text": "python3 - <<'PY'\nfrom pathlib import Path\nimport yaml\nroot=Path('/home/avifenesh/projects/aviary')\ntotal=0\nfor p in sorted((root/'data/evals').glob('*/scorecards.yml')):\n dat...",
"tool_name": "exec_command"
},
"label": 1,
"source_event_id": "t15",
"support_probability": 0.9928786158561707
}
]
}
]
},
"train": "data/tmp/mixed_mode_train.v2.field_decision.jsonl",
"train_candidate_pairs": 63752,
"train_groups": 5882,
"train_rows": 1217
}
|