Datasets:
File size: 75,246 Bytes
b8ea052 | 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 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 | """Deterministic lowering from reviewed predicate IR to executable programs.
The natural-language layer may propose a candidate, but this module is the
only promotion path for the six reviewed contract families. It preserves
every parameter and region, selects only registered materializers, and emits
synthetic witnesses proving that decision parameters are live.
"""
from __future__ import annotations
from copy import deepcopy
import re
from typing import Any, Mapping, Sequence
from pvr_eval.contracts.change_spec import compile_change_spec
from pvr_eval.contracts.contract_obligations import (
build_binding_witnesses,
compile_parameter_bindings,
)
from pvr_eval.contracts.fact_producer_registry import (
objective_obligation_producer,
objective_producer_for_binding,
)
from pvr_eval.contracts.metric_profiles import (
ARCHITECTURE_QUALITATIVE_FAILURE_CERTIFICATE_VERSION,
ARTICULATED_ENTITY_PRESERVATION_EXECUTOR_VERSION,
ARTICULATED_ENTITY_PRESERVATION_PROFILE,
HUMAN_POSE_PROFILE,
LOCAL_CORRESPONDENCE_PROFILE,
REGISTERED_ARCHITECTURE_PRESERVATION_EXECUTOR_VERSION,
REGISTERED_ARCHITECTURE_PRESERVATION_PROFILE,
articulated_entity_preservation_profile_parameters,
registered_architecture_preservation_profile_parameters,
registered_grid_structure_profile_parameters,
support_contact_geometry_profile_parameters,
)
from pvr_eval.contracts.predicate_ir_candidate import (
FAMILY_OPERATION_TEMPLATES,
RULE_FAMILY_TARGETS,
compile_predicate_ir_candidate,
validate_predicate_ir_candidate,
)
LOWERED_FAMILIES = frozenset(FAMILY_OPERATION_TEMPLATES)
COMPOSITE_FAMILY = "contract_family_all_of"
LOWERER_VERSION = "pvr.reviewed_contract_lowerer.v12"
ALLOWED_EDIT_QUERY_SCOPE_POLICY = "reviewed_exclusion_clause_subject_intersection_v1"
ROSTER_ALLOWED_EDIT_QUERY_SCOPE_POLICY = "question_support_roster_queries_v1"
VERIFIED_GROUP_ROSTER_PROFILE = "verified_group_roster_assignment_v1"
# A source rule name alone is not a type. These signatures prevent an LLM or
# stale resolver from attaching unrelated parameters to a registered family
# and receiving an executable status merely because the name is recognized.
RULE_REQUIRED_PARAMETERS: dict[str, frozenset[str]] = {
"multi_region_strict_preservation": frozenset({"maximum_changed_pixel_fraction", "minimum_region_ssim"}),
"architecture_and_decor_preservation": frozenset({"registration_residual_tolerance_pixels", "minimum_edge_preservation"}),
"masked_entity_structure_preservation": frozenset({"minimum_major_part_match_rate", "major_parts"}),
"instance_and_background_preservation": frozenset({"required_original_bird_match_rate", "maximum_background_changed_fraction"}),
"masked_object_pose_and_appearance_preservation": frozenset({"minimum_robot_mask_iou", "maximum_centroid_shift_fraction_of_image_diagonal"}),
"masked_environment_structure_preservation": frozenset({"minimum_invariant_region_ssim", "maximum_unexplained_change_area_fraction"}),
"matched_instance_integrity": frozenset({"appearance_similarity_warning_threshold", "mask_shape_change_warning_threshold"}),
"protected_subject_invariance": frozenset({"centroid_shift_max", "bbox_scale_ratio"}),
"registered_structure_preservation": frozenset({"maximum_normalized_boundary_displacement", "compare_drawer_state"}),
"multi_instance_furniture_preservation": frozenset({"minimum_mask_iou_after_registration", "require_equal_chair_count"}),
"registered_object_pose_preservation": frozenset({"registered_mask_iou_min", "expected_instance_count"}),
"all_landmarks_preserved": frozenset({"registered_mask_iou_min", "require_one_to_one_instance_matching"}),
"movable_inventory_integrity": frozenset({"expected_counts", "appearance_embedding_similarity_min"}),
"all_instances_strict_preservation": frozenset({"minimum_mask_iou_after_registration", "expected_magazine_count"}),
"strict_person_instance_preservation": frozenset({"minimum_mask_iou_after_registration", "minimum_appearance_similarity"}),
"all_furniture_structures_preserved": frozenset({"minimum_boundary_iou", "minimum_unaffected_feature_match_ratio"}),
"architectural_structure_preservation": frozenset({"minimum_door_mask_iou", "minimum_wall_feature_similarity"}),
"multi_region_invariant_preservation": frozenset({"minimum_man_mask_iou", "minimum_board_feature_inlier_ratio"}),
"structure_and_tray_preservation": frozenset({"minimum_structural_line_correspondence", "minimum_tray_shape_iou"}),
"relocated_object_in_air_zone": frozenset({"minimum_air_zone_containment", "maximum_solid_mask_overlap_for_pass"}),
"categorized_group_completeness": frozenset({"expected_counts", "minimum_group_members_for_clear_group"}),
"dual_category_grouping": frozenset({"personal_care_expected", "medical_expected"}),
"document_group_completeness": frozenset({"expected_counts", "use_title_or_color_identity"}),
"remote_pair_on_table_surface": frozenset({"required_instances", "minimum_mask_containment"}),
"paired_relocation_beside_dispenser": frozenset({"destination_reference", "require_cart_and_chair"}),
"same_unit_different_compartment_placement": frozenset({"pass_min_containment", "pass_max_tray_overlap"}),
"supported_nonintersecting_placement": frozenset({
"pass_max_support_gap_ratio",
"partial_max_support_gap_ratio",
"pass_min_supported_bottom_fraction",
"fail_below_supported_bottom_fraction",
"pass_max_structural_overlap",
"partial_max_structural_overlap",
}),
"open_shelf_state_preserved": frozenset({"reference_state", "minimum_visible_recess_fraction"}),
"flat_tabletop_structure_preserved": frozenset({"forbidden_surface_features", "minimum_forbidden_feature_confidence"}),
"upright_group_arrangement": frozenset({"upright_angle_tolerance_degrees", "pass_min_upright_fraction"}),
}
def _mapping(value: Any) -> dict[str, Any]:
return deepcopy(dict(value)) if isinstance(value, Mapping) else {}
def _list(value: Any) -> list[Any]:
return list(value) if isinstance(value, (list, tuple)) else []
def _dedupe(values: Sequence[Any]) -> list[str]:
result: list[str] = []
seen: set[str] = set()
for item in values:
value = str(item or "").strip()
if value and value not in seen:
seen.add(value)
result.append(value)
return result
def _configure_verified_group_roster(
parameters: dict[str, Any], reasons: list[str],
) -> None:
"""Promote a complete reviewed roster to shared group matching runtime config."""
annotations = [
item for item in _list(parameters.get("source_instance_annotations"))
if isinstance(item, Mapping)
]
expected_raw = parameters.get("source_roster_expected_counts")
if not annotations or not isinstance(expected_raw, Mapping):
return
observed: dict[str, int] = {}
instance_ids: set[str] = set()
labels: list[str] = []
animal_terms = {
"animal", "bird", "birds", "cat", "deer", "doe", "dog", "fawn",
"fox", "raccoon", "squirrel", "wildlife",
}
animal_roster = True
for index, annotation in enumerate(annotations):
group = str(annotation.get("group_id") or "").strip()
instance_id = str(annotation.get("instance_id") or "").strip()
if not group:
reasons.append(f"verified_source_roster_group_missing:{index}")
continue
if not instance_id or instance_id in instance_ids:
reasons.append(f"verified_source_roster_instance_id_invalid:{index}")
instance_ids.add(instance_id)
observed[group] = observed.get(group, 0) + 1
label = str(annotation.get("label") or group).strip()
if label.endswith("s") and not label.endswith("ss"):
label = label[:-1]
if label and label not in labels:
labels.append(label)
animal_roster = animal_roster and bool(
set(re.findall(r"[a-z]+", f"{group} {label}".casefold())) & animal_terms
)
try:
expected = {str(group): int(count) for group, count in expected_raw.items()}
except (TypeError, ValueError):
reasons.append("verified_source_roster_expected_counts_invalid")
return
if expected != observed:
reasons.append("verified_source_roster_expected_counts_mismatch")
return
if parameters.get("source_roster_complete") is not True:
reasons.append("verified_source_roster_not_declared_complete")
return
if str(parameters.get("source_inventory_provenance") or "") not in {
"contract_source_annotations", "manual_source_annotations", "certified_source_inventory",
}:
reasons.append("verified_source_roster_provenance_uncertified")
return
distractor = "other animal" if animal_roster else "other object"
parameters.setdefault("source_annotation_mask_profile", "fastsam_bbox_prompt_or_color_component_v2")
parameters.setdefault("matching_classification_labels", [*labels, distractor])
parameters.setdefault("matching_use_segmenter_proposals", True)
parameters.setdefault("matching_segmenter_min_area_fraction", 0.00003)
parameters.setdefault("matching_segmenter_max_area_fraction", 0.08)
parameters.setdefault("matching_segmenter_max_proposals", 160)
parameters.setdefault("matching_classification_minimum_margin", 0.0)
parameters.setdefault("matching_classification_minimum_probability", 0.25)
parameters.setdefault("minimum_match_confidence", 0.55)
parameters.setdefault(
"required_matching_features",
["verified group label", "appearance embedding", "relative size", "mask shape"],
)
parameters.setdefault("verified_roster_profile", VERIFIED_GROUP_ROSTER_PROFILE)
question_support = parameters.get("question_support_roster")
question_support_purpose = (
str(question_support.get("purpose") or "")
if isinstance(question_support, Mapping)
else ""
)
parameters.setdefault(
"verified_roster_assignment_mode",
(
"question_support_allowed_edit_coverage_v1"
if "allowed_edit" in question_support_purpose
else "identity_exact"
),
)
parameters.setdefault("verified_roster_expected_counts", expected)
parameters.setdefault("verified_roster_minimum_area_ratio", 0.10)
parameters.setdefault("verified_roster_maximum_area_ratio", 6.0)
_QUERY_STOPWORDS = frozenset(
{
"all", "and", "central", "every", "input", "output", "the", "with",
}
)
_QUERY_PREPOSITIONS = frozenset(
{"at", "beside", "by", "from", "in", "inside", "near", "on", "onto", "under", "with"}
)
_QUERY_STATE_WORDS = frozenset({"lying", "perched", "relocated", "resting", "standing", "upright"})
def _query_token(value: str) -> str:
token = value.lower()
return token[:-1] if token.endswith("s") and len(token) > 3 else token
def _semantic_query_head(query: str) -> str:
tokens = [
token for token in re.findall(r"[a-z0-9]+", query.lower())
if token not in _QUERY_STOPWORDS
]
for index, token in enumerate(tokens):
if token in _QUERY_PREPOSITIONS:
tokens = tokens[:index]
break
while tokens and tokens[-1] in _QUERY_STATE_WORDS:
tokens.pop()
return _query_token(tokens[-1]) if tokens else ""
def _explicit_mask_heads(text: str) -> set[str]:
"""Return entity words explicitly used to construct an edit mask."""
heads: set[str] = set()
for match in re.finditer(
r"([a-z0-9]+(?:[ -]+[a-z0-9]+){0,4})[ -]+masks?\b",
text.lower(),
):
tokens = [
token for token in re.findall(r"[a-z0-9]+", match.group(1))
if token not in _QUERY_STOPWORDS
and token not in {
"allowed", "destination", "edit", "input", "matched", "movable",
"original", "output", "relocated", "source",
}
]
heads.update(_query_token(token) for token in tokens[-2:])
if re.search(r"\b(?:its|their)\s+masks?\b", text.lower()):
for match in re.finditer(
r"\bsegment\s+(?:the\s+)?(.+?)(?=\s+(?:and|from|in|inside|near|on|perched|with)\b|[,;.])",
text.lower(),
):
action_tokens = [
token for token in re.findall(r"[a-z0-9]+", match.group(1))
if token not in _QUERY_STOPWORDS and token not in {"every"}
]
if action_tokens:
heads.add(_query_token(action_tokens[-1]))
if "animal" in heads:
heads.update({"wildlife", "bird"})
if "wildlife" in heads:
heads.add("animal")
return heads
def _allowed_edit_queries(
region: Mapping[str, Any],
*,
subjects: Sequence[str],
related_regions: Sequence[Mapping[str, Any]],
) -> list[str]:
"""Select only subject queries explicitly named as edit exclusions.
Reviewed checks often use one broad ``object_queries`` list for both the
protected scene and the editable foreground. Copying that whole list to
an allowed-edit mask can hide real damage (for example by subtracting the
forest ground and rocks). The editable query set is therefore derived
from the exclusion clauses themselves. If no reviewed subject matches,
the region remains unsupported instead of inventing a foreground class.
"""
texts = [str(region.get("query_or_rule") or "")]
for other in related_regions:
if other is region:
continue
rule = str(other.get("query_or_rule") or "")
if re.search(r"\b(exclud\w*|minus|subtract\w*)\b", rule.lower()):
texts.append(rule.split(";", 1)[0])
explicit_heads: set[str] = set()
for text in texts:
explicit_heads.update(_explicit_mask_heads(text))
reviewed_support = _dedupe(_list(region.get("reviewed_support_queries")))
candidates = reviewed_support or _dedupe(subjects)
selected: list[str] = []
for subject in candidates:
# Only a noun explicitly attached to a reviewed mask is eligible.
# Merely appearing as an anchor ("bird on robot shoulder") must not
# make the robot itself editable.
head = _semantic_query_head(subject)
reviewed_tokens = {
_query_token(token)
for token in re.findall(r"[a-z0-9]+", subject.lower())
if token not in _QUERY_STOPWORDS
}
if (
head in explicit_heads
or bool(reviewed_support and reviewed_tokens.intersection(explicit_heads))
):
selected.append(subject)
return selected
def _scope_allowed_edit_regions(
regions: list[dict[str, Any]], subjects: Sequence[str], *, roster_queries: Sequence[str] = (),
) -> None:
"""Remove protected/context queries from detector-backed edit masks."""
for region in regions:
if str(region.get("semantic_role") or "") != "allowed_edit_exclusion":
continue
materializer = region.get("materializer")
if not isinstance(materializer, dict):
continue
if str(materializer.get("kind") or "") != "detector_segmenter":
continue
if roster_queries:
materializer["queries"] = _dedupe(roster_queries)
materializer["query_scope_policy"] = ROSTER_ALLOWED_EDIT_QUERY_SCOPE_POLICY
else:
materializer["queries"] = _allowed_edit_queries(
region, subjects=subjects, related_regions=regions,
)
materializer["query_scope_policy"] = ALLOWED_EDIT_QUERY_SCOPE_POLICY
def _exclusion_parameter_requested(parameters: Mapping[str, Any]) -> bool:
"""Return whether a reviewed config explicitly requires an exclusion."""
# Occlusion can be a destination-relation fact (for example a chair
# legitimately hidden behind a cart) without authorizing pixels to be
# removed from a preservation comparison. Only mask/exemption wording
# that actually requests subtraction activates the exclusion gate.
return any(
(
str(key).casefold().startswith("exclude_")
or "exemption" in str(key).casefold()
or "allowed_mask" in str(key).casefold()
or "occlusion_mask" in str(key).casefold()
)
and value not in (False, None, "", [], {})
for key, value in parameters.items()
)
def _append_parameter_allowed_edit_region(
regions: list[dict[str, Any]], parameters: Mapping[str, Any], reasons: list[str],
*, require_complete: bool,
) -> None:
"""Lower an exclusion config only from explicit reviewed object queries.
A boolean such as ``exclude_verified_new_furniture_occlusion`` says that
an exclusion is required, but it does not identify the pixels. The
objective executor therefore requires a separate ``allowed_edit_queries``
list and refuses to invent one from protected subjects or sibling
evidence. Semantic-fallback-only rules remain fallback candidates.
"""
if not _exclusion_parameter_requested(parameters):
return
if any(region.get("semantic_role") == "allowed_edit_exclusion" for region in regions):
return
queries = _dedupe(_list(parameters.get("allowed_edit_queries")))
if not queries:
if require_complete:
reasons.append("exclusion_operation_missing_allowed_edit_region_or_queries")
return
mask_phrases = [f"{query} masks" for query in queries]
regions.append(
{
"region_id": "contract_allowed_edit_exclusion",
"semantic_role": "allowed_edit_exclusion",
"image_roles": ["pred"],
"source_image": "pred",
"proposal_method": "sam_mask",
"query_or_rule": (
"Union " + " and ".join(mask_phrases)
+ " as the allowed output occlusion."
),
"reviewed_support_queries": queries,
"materializer": {
"kind": "detector_segmenter",
"queries": queries,
"selection_policy": "reviewed_query_topk_then_sam",
"cardinality": "contract_defined",
"region_rule": (
"Union " + " and ".join(mask_phrases)
+ " as the allowed output occlusion."
),
},
"geometry_policy": "mask_or_polygon_required",
"provenance": "derived from reviewed allowed_edit_queries operation config",
}
)
def _configure_multi_region_protected_regions(
regions: list[dict[str, Any]], source_rule_type: str,
) -> None:
"""Make every named invariant ROI measurable for strict multi-region rules."""
if source_rule_type != "multi_region_strict_preservation":
return
for region in regions:
if region.get("semantic_role") in {
"allowed_edit_exclusion", "comparison_context_exclusion",
}:
continue
region["semantic_role"] = "protected_region"
region["image_roles"] = ["input", "pred"]
region["source_image"] = "both"
def _configure_registered_architecture_regions(
regions: list[dict[str, Any]], parameters: Mapping[str, Any], reasons: list[str],
) -> None:
"""Compile one architecture/decor contract into strict component masks."""
component_queries = _dedupe(_list(parameters.get("architecture_component_queries")))
decor_queries = _dedupe(_list(parameters.get("tidy_decor_queries")))
allowed_queries = _dedupe(_list(parameters.get("allowed_edit_queries")))
for key, values in (
("architecture_component_queries", component_queries),
("tidy_decor_queries", decor_queries),
("allowed_edit_queries", allowed_queries),
):
if not values:
reasons.append(f"registered_architecture_{key}_missing")
configured_roles: set[str] = set()
for region in regions:
region_id = str(region.get("region_id") or "")
semantic_role = str(region.get("semantic_role") or "")
lowered_id = region_id.casefold()
component_role = ""
queries: list[str] = []
if semantic_role == "allowed_edit_exclusion" or "allowed" in lowered_id:
component_role, queries = "allowed_edit", allowed_queries
region["semantic_role"] = "allowed_edit_exclusion"
elif "decor" in lowered_id:
component_role, queries = "tidy_decor", decor_queries
region["semantic_role"] = "context_region"
elif semantic_role == "protected_region" or any(
token in lowered_id for token in ("architecture", "structure", "invariant")
):
component_role, queries = "major_structure", component_queries
region["semantic_role"] = "protected_region"
if not component_role:
continue
configured_roles.add(component_role)
region["image_roles"] = ["input", "pred"]
region["source_image"] = "both"
region["materializer"] = {
"kind": "architecture_region_parser",
"profile": REGISTERED_ARCHITECTURE_PRESERVATION_PROFILE,
"component_role": component_role,
"queries": queries,
"strict_component_masks": True,
"selection_policy": "one_reviewed_query_top1_then_strict_mask",
"geometry_policy": "strict_mask_union_with_component_certificate",
"region_rule": str(region.get("query_or_rule") or ""),
}
region["geometry_policy"] = "strict_mask_union_with_component_certificate"
for role in ("major_structure", "tidy_decor", "allowed_edit"):
if role not in configured_roles:
reasons.append(f"registered_architecture_region_missing:{role}")
def _append_explicit_comparison_context_exclusion(
regions: list[dict[str, Any]], *, subjects: Sequence[str], roster_queries: Sequence[str],
) -> None:
"""Materialize non-question foreground explicitly excluded by the contract.
For background-preservation questions, a fixed foreground anchor such as a
robot may be outside the question while still being named in the reviewed
``full frame minus ... mask`` rule. It is a comparison-context exclusion,
never an editable roster member.
"""
heads: set[str] = set()
for region in regions:
if str(region.get("semantic_role") or "") != "protected_region":
continue
rule = str(region.get("query_or_rule") or "")
if re.search(r"\b(exclud\w*|minus|subtract\w*)\b", rule.lower()):
heads.update(_explicit_mask_heads(rule))
roster_heads = {_semantic_query_head(query) for query in roster_queries}
if "animal" in roster_heads or any(
re.search(r"\b(?:animal|bird|deer|doe|fawn|fox|raccoon|squirrel|wildlife)\b", query.lower())
for query in roster_queries
):
roster_heads.update({"animal", "wildlife"})
queries = [
subject for subject in _dedupe(subjects)
if _semantic_query_head(subject) in heads
and _semantic_query_head(subject) not in roster_heads
]
if not queries:
return
regions.append(
{
"region_id": "question_context_exclusion",
"semantic_role": "comparison_context_exclusion",
"image_roles": ["input", "pred"],
"source_image": "both",
"proposal_method": "derived_from_reviewed_exclusion_clause",
"query_or_rule": "Exclude only explicitly named non-question foreground from background comparison.",
"materializer": {
"kind": "detector_segmenter",
"queries": queries,
"selection_policy": "reviewed_query_topk_then_sam",
"cardinality": "contract_defined",
"query_scope_policy": "reviewed_comparison_context_exclusion_v1",
},
"geometry_policy": "mask_or_polygon_required",
"provenance": "derived from reviewed protected-region exclusion clause",
}
)
def reviewed_region_policy_reasons(predicate: Mapping[str, Any]) -> list[str]:
"""Validate lowering policies whose semantics are not covered by JSON shape.
A previously compiled predicate can remain schema-valid after the lowerer is
fixed. In particular, old allowed-edit masks copied every object query and
could subtract protected background from the preservation comparison. The
compilation gate calls this validator so such stale artifacts cannot enter
a new run silently.
"""
provenance = _mapping(predicate.get("provenance"))
if provenance.get("compiler") != "pvr_eval.contracts.predicate_lowering.lower_reviewed_predicate":
return []
regions = [
deepcopy(dict(item))
for item in _list(predicate.get("regions"))
if isinstance(item, Mapping)
]
subjects = _dedupe(
str(item.get("identity_query") or "")
for item in _list(predicate.get("subjects"))
if isinstance(item, Mapping)
)
parameters = _mapping(predicate.get("parameters"))
roster_queries = _dedupe(
str(item.get("query_text") or "")
for item in _list(parameters.get("source_instance_annotations"))
if isinstance(item, Mapping)
)
reasons: list[str] = []
capabilities = predicate.get("capability_versions") if isinstance(predicate.get("capability_versions"), Mapping) else {}
if capabilities.get("lowerer") != LOWERER_VERSION:
reasons.append("stale_reviewed_lowerer_version")
if (
parameters.get("articulated_entity_preservation_profile")
== ARTICULATED_ENTITY_PRESERVATION_PROFILE
and parameters.get("articulated_entity_preservation_executor_version")
!= ARTICULATED_ENTITY_PRESERVATION_EXECUTOR_VERSION
):
reasons.append("stale_articulated_entity_preservation_executor_version")
if parameters.get("registered_architecture_profile") == REGISTERED_ARCHITECTURE_PRESERVATION_PROFILE:
if parameters.get("registered_architecture_executor_version") != (
REGISTERED_ARCHITECTURE_PRESERVATION_EXECUTOR_VERSION
):
reasons.append("stale_registered_architecture_executor_version")
expected_by_role = {
"major_structure": _dedupe(_list(parameters.get("architecture_component_queries"))),
"tidy_decor": _dedupe(_list(parameters.get("tidy_decor_queries"))),
"allowed_edit": _dedupe(_list(parameters.get("allowed_edit_queries"))),
}
seen_roles: set[str] = set()
for region in regions:
materializer = _mapping(region.get("materializer"))
component_role = str(materializer.get("component_role") or "")
if component_role not in expected_by_role:
continue
seen_roles.add(component_role)
if materializer.get("kind") != "architecture_region_parser":
reasons.append(f"registered_architecture_materializer_mismatch:{component_role}")
if materializer.get("strict_component_masks") is not True:
reasons.append(f"registered_architecture_strict_mask_policy_missing:{component_role}")
if _dedupe(_list(materializer.get("queries"))) != expected_by_role[component_role]:
reasons.append(f"registered_architecture_query_scope_mismatch:{component_role}")
for component_role, expected in expected_by_role.items():
if not expected or component_role not in seen_roles:
reasons.append(f"registered_architecture_component_role_missing:{component_role}")
if any(
str(region.get("semantic_role") or "") == "comparison_context_exclusion"
for region in regions
):
reasons.append("registered_architecture_decor_must_not_be_comparison_exclusion")
if (
objective_obligation_producer(
str(predicate.get("family") or ""),
str(predicate.get("source_rule_type") or ""),
)
and _exclusion_parameter_requested(parameters)
and not any(
str(region.get("semantic_role") or "") == "allowed_edit_exclusion"
for region in regions
)
):
reasons.append("exclusion_operation_missing_allowed_edit_region")
for region in regions:
if str(region.get("semantic_role") or "") != "allowed_edit_exclusion":
continue
materializer = _mapping(region.get("materializer"))
if str(materializer.get("kind") or "") != "detector_segmenter":
continue
region_id = str(region.get("region_id") or "missing")
expected_policy = (
ROSTER_ALLOWED_EDIT_QUERY_SCOPE_POLICY
if roster_queries else ALLOWED_EDIT_QUERY_SCOPE_POLICY
)
if materializer.get("query_scope_policy") != expected_policy:
reasons.append(f"stale_allowed_edit_query_scope_policy:{region_id}")
continue
expected = roster_queries or _allowed_edit_queries(
region, subjects=subjects, related_regions=regions,
)
actual = _dedupe(_list(materializer.get("queries")))
if actual != expected:
reasons.append(f"allowed_edit_query_scope_mismatch:{region_id}")
return reasons
def reviewed_source_check_projection(check: Mapping[str, Any]) -> dict[str, Any]:
"""Reconstruct the exact check hashed before compiler artifacts were embedded."""
source = deepcopy(dict(check))
source.pop("predicate_ir_candidate", None)
source.pop("executable_predicate", None)
params = _mapping(source.get("params"))
params.pop("predicate_ir_candidate", None)
params.pop("executable_predicate", None)
params.pop("require_executable_predicate", None)
source["params"] = params
return source
def reviewed_source_check_sha256(check: Mapping[str, Any]) -> str:
"""Return the canonical hash used by reviewed predicate artifacts."""
candidate = compile_predicate_ir_candidate(reviewed_source_check_projection(check))
return str(candidate.get("source_check_sha256") or "")
def _roles(region: Mapping[str, Any]) -> list[str]:
explicit = _list(region.get("image_roles"))
if explicit:
return _dedupe(explicit)
source = str(region.get("source_image") or region.get("image_role") or "pred").lower()
if source in {"both", "input_pred", "input/output"}:
return ["input", "pred"]
if source in {"output", "prediction", "pred"}:
return ["pred"]
return ["input"]
def _special_vlm_replacement(region_id: str) -> tuple[str, list[str], str]:
"""Replace the four reviewed VLM boxes with deterministic backends."""
if region_id == "airborne_flock_zone":
return "derived_free_space_zone", ["robot", "ground", "tree trunk", "tree branch"], "air_above_solid_exclusions"
if region_id == "central_walkway":
return "planar_free_space_parser", ["rug", "open floor", "sofa", "TV stand"], "floor_corridor_between_furniture"
if region_id == "map_pinboard":
return "detector_segmenter", ["wall pinboard", "map board with pinned notes"], "top1_then_segment"
if region_id == "house_structure_mask":
return "architecture_region_parser", ["house walls", "windows", "eaves", "building facade"], "fixed_architecture_union"
return "", [], ""
def _materialize_region(
raw: Mapping[str, Any],
*,
subjects: Sequence[str],
) -> tuple[dict[str, Any], list[str]]:
region = deepcopy(dict(raw))
region_id = str(region.get("region_id") or region.get("id") or "").strip()
method = str(region.get("proposal_method") or "").strip()
query_or_rule = str(region.get("query_or_rule") or "").strip()
roles = _roles(region)
reasons: list[str] = []
queries: list[str] = []
raw_materializer = (
region.get("materializer")
if isinstance(region.get("materializer"), Mapping)
else {}
)
reviewed_support_queries = _dedupe(
_list(
region.get("object_queries")
or region.get("queries")
or region.get("reviewed_support_queries")
or raw_materializer.get("queries")
)
)
materializer: dict[str, Any]
if method in {"grounding_dino_box", "dino_box"}:
queries = reviewed_support_queries or _dedupe(subjects)
materializer = {
"kind": "detector_segmenter",
"queries": queries,
"selection_policy": "reviewed_query_topk_then_sam",
"cardinality": "contract_defined",
}
elif method == "sam_mask":
# SAM is never invoked from prose alone. The reviewed subject queries
# are its explicit detector seed; the region rule only filters/merges.
queries = reviewed_support_queries or _dedupe(subjects)
materializer = {
"kind": "detector_segmenter",
"queries": queries,
"seed": "check_scoped_detector_boxes",
"selection_policy": "detector_seed_then_rule_filtered_sam",
"region_rule": query_or_rule,
}
if not queries:
reasons.append(f"region_seed_queries_missing:{region_id}")
elif method == "change_mask":
materializer = {
"kind": "registered_change_derivation",
"registration": "robust_static_background_alignment",
"operation": "union_subtract_clip",
"region_rule": query_or_rule,
}
elif method == "ocr_layout":
materializer = {
"kind": "ocr_container_layout",
"ocr_scope": "full_frame_then_nearest_container",
"region_rule": query_or_rule,
}
elif method == "existing_zone":
materializer = {
"kind": "anchor_relative_zone",
"queries": ["water dispenser"],
"geometry_policy": "anchor_side_polygon_with_forbidden_subtraction",
"region_rule": query_or_rule,
}
elif method == "grid_parser":
materializer = {
"kind": "shelf_grid_parser",
"queries": reviewed_support_queries or ["shelving unit", "shelf compartment"],
"geometry_policy": "registered_rectified_compartment_grid",
"region_rule": query_or_rule,
}
elif method == "vlm_region_box":
kind, queries, policy = _special_vlm_replacement(region_id)
if not kind:
reasons.append(f"zero_vlm_materializer_missing:{region_id}")
materializer = {"kind": "unsupported"}
else:
materializer = {
"kind": kind,
"queries": queries,
"geometry_policy": policy,
"region_rule": query_or_rule,
}
else:
reasons.append(f"unregistered_region_materializer:{region_id}:{method or 'missing'}")
materializer = {"kind": "unsupported"}
semantic_role = "context_region"
lowered_id = region_id.lower()
if any(token in lowered_id for token in ("allowed", "edit", "exemption")):
semantic_role = "allowed_edit_exclusion"
elif any(token in lowered_id for token in ("destination", "target_compartment", "storage", "airborne", "table_top")):
semantic_role = "destination_region"
elif any(token in lowered_id for token in ("invariant", "preservation", "structure", "surface")):
semantic_role = "protected_region"
lowered = {
"region_id": region_id,
"semantic_role": semantic_role,
"image_roles": roles,
"source_image": "both" if set(roles) == {"input", "pred"} else roles[0],
"proposal_method": method,
"query_or_rule": query_or_rule,
"materializer": materializer,
"geometry_policy": str(materializer.get("geometry_policy") or "mask_or_polygon_required"),
"provenance": "reviewed planned_regions",
}
if reviewed_support_queries:
lowered["reviewed_support_queries"] = reviewed_support_queries
return lowered, reasons
def _configure_registered_grid_structure_regions(
regions: list[dict[str, Any]], subjects: Sequence[str],
) -> None:
"""Route a reviewed shelf structure through the shared grid parser."""
shelf_queries = [
value for value in _dedupe(subjects)
if "shel" in value.casefold() and "unit" in value.casefold()
]
if not shelf_queries:
shelf_queries = [
value for value in _dedupe(subjects)
if "shel" in value.casefold()
][:1]
for region in regions:
if str(region.get("region_id") or "") != "shelving_unit_structure":
continue
region["image_roles"] = ["input", "pred"]
region["source_image"] = "both"
region["materializer"] = {
"kind": "shelf_grid_parser",
"queries": shelf_queries or ["shelving unit"],
"geometry_policy": "registered_rectified_compartment_grid",
"region_rule": str(region.get("query_or_rule") or ""),
"profile": "registered_grid_structure_preservation_v1",
}
region["geometry_policy"] = "registered_rectified_compartment_grid"
def _operation(
operation_id: str,
kind: str,
tool_family: str,
image_roles: Sequence[str],
*,
depends_on: Sequence[str] = (),
produces: Sequence[str] = (),
) -> dict[str, Any]:
return {
"id": operation_id,
"kind": kind,
"tool_family": tool_family,
"image_roles": list(image_roles),
"depends_on": list(depends_on),
"parameters": {},
"produces": list(produces),
}
def _operations(families: Sequence[str], parameters: Mapping[str, Any] | None = None) -> list[dict[str, Any]]:
parameters = parameters or {}
local_facts_requested = any(
key in parameters
for key in ("max_rotation_change_degrees", "maximum_rotation_change_degrees", "minimum_local_feature_match_ratio")
) and bool(parameters.get("local_correspondence_profile"))
needs_fixed = "fixed_location_entity_preservation" in families
articulated_profile = (
str(parameters.get("articulated_entity_preservation_profile") or "")
== ARTICULATED_ENTITY_PRESERVATION_PROFILE
)
needs_region = articulated_profile or any(family in families for family in (
"registered_invariant_region_preservation", "destination_membership", "registered_structural_state",
))
needs_match = any(family in families for family in (
"fixed_location_entity_preservation", "relocated_entity_integrity", "destination_membership",
)) or bool(parameters.get("source_instance_annotations")) or local_facts_requested
needs_registration = needs_fixed or any(family in families for family in (
"registered_invariant_region_preservation", "registered_structural_state",
)) or local_facts_requested
needs_change = articulated_profile or "registered_invariant_region_preservation" in families
needs_state = any(family in families for family in (
"destination_membership", "registered_structural_state", "group_pose_arrangement",
))
needs_attribute = any(family in families for family in (
"fixed_location_entity_preservation", "relocated_entity_integrity", "group_pose_arrangement",
))
operations = [
_operation("detect_input", "object_detection", "detector", ("input",), produces=("input_boxes",)),
_operation("detect_pred", "object_detection", "detector", ("pred",), produces=("pred_boxes",)),
_operation("segment_input", "mask_refinement", "segmenter", ("input",), depends_on=("detect_input",), produces=("input_masks",)),
_operation("segment_pred", "mask_refinement", "segmenter", ("pred",), depends_on=("detect_pred",), produces=("pred_masks",)),
]
terminal: list[str] = ["segment_input", "segment_pred"]
if needs_match:
operations.append(_operation(
"match_instances", "instance_matching", "matcher", ("input", "pred"),
depends_on=("segment_input", "segment_pred"), produces=("instance_matches",),
))
terminal.append("match_instances")
if needs_registration:
operations.append(_operation(
"register_frames", "image_registration", "registration", ("input", "pred"),
produces=("registered_transform", "registration_quality"),
))
terminal.append("register_frames")
if needs_region:
region_deps = ("register_frames",) if needs_registration else ()
operations.append(_operation(
"materialize_regions", "region_resolution", "typed_region_materializer", ("input", "pred"),
depends_on=region_deps, produces=("typed_regions", "materializer_status"),
))
operations.append(_operation(
"refine_region_geometry", "spatial_geometry", "segmenter", ("input", "pred"),
depends_on=("materialize_regions",), produces=("region_masks", "zone_geometry"),
))
terminal.append("refine_region_geometry")
if needs_change:
deps = ["register_frames"]
if needs_region:
deps.append("refine_region_geometry")
operations.append(_operation(
"measure_registered_change", "change_analysis", "change_analyzer", ("input", "pred"),
depends_on=deps, produces=("change_metrics", "structure_metrics"),
))
terminal.append("measure_registered_change")
if needs_state:
deps = ["segment_pred"] + (["refine_region_geometry"] if needs_region else [])
operations.append(_operation(
"parse_visual_state", "state_parsing", "state_parser", ("input", "pred"),
depends_on=deps, produces=("state_facts", "pose_facts", "layout_facts"),
))
terminal.append("parse_visual_state")
if "destination_membership" in families:
operations.append(_operation(
"count_destination_instances", "counting", "checker_internal", ("pred",),
depends_on=("segment_pred",), produces=("count_facts",),
))
terminal.append("count_destination_instances")
if needs_attribute:
operations.append(_operation(
"measure_visual_attributes", "visual_attribute", "attribute_parser", ("input", "pred"),
depends_on=("segment_input", "segment_pred"), produces=("attribute_facts", "appearance_facts"),
))
terminal.append("measure_visual_attributes")
if "maximum_normalized_keypoint_error" in parameters:
operations.append(_operation(
"measure_human_pose", "pose_estimation", "human_pose_parser", ("input", "pred"),
depends_on=("detect_input", "detect_pred"), produces=("human_pose_keypoints",),
))
terminal.append("measure_human_pose")
if local_facts_requested:
operations.append(_operation(
"measure_local_correspondence", "local_correspondence", "local_correspondence", ("input", "pred"),
depends_on=("match_instances", "register_frames"),
produces=("rotation_change_degrees", "local_feature_match_ratio", "local_correspondence_certificate"),
))
terminal.append("measure_local_correspondence")
if parameters.get("source_selector_profile"):
operations.append(_operation(
"select_allowed_edit_source", "source_instance_selection",
"source_instance_selector", ("input",),
depends_on=("segment_input",), produces=("source_selector_certificate",),
))
terminal.append("select_allowed_edit_source")
operations.append(_operation(
"decide", "predicate_decision", "contract_metric_binding", ("input", "pred"),
depends_on=tuple(dict.fromkeys(terminal)), produces=("check_result",),
))
return operations
def _subject_spec(
index: int,
aliases: Sequence[str],
*,
expected_count: int = 1,
provenance: str = "reviewed object_queries",
subject_id: str = "",
component_queries: Sequence[str] = (),
part_schema: Mapping[str, Any] | None = None,
) -> dict[str, Any]:
clean_aliases = _dedupe(aliases)
spec: dict[str, Any] = {
"id": subject_id or f"subject_{index + 1}",
"identity_query": clean_aliases[0],
"input_queries": clean_aliases,
"output_queries": clean_aliases,
"aliases": clean_aliases,
"expected_count": max(1, int(expected_count)),
"cardinality_mode": "exact",
"provenance": provenance,
}
clean_components = _dedupe(component_queries)
if clean_components:
spec["component_queries"] = clean_components
if isinstance(part_schema, Mapping) and part_schema:
spec["part_schema"] = deepcopy(dict(part_schema))
return spec
def _explicit_physical_subject_specs(
subjects: Sequence[str], parameters: Mapping[str, Any],
) -> tuple[list[dict[str, Any]], list[str]] | None:
"""Compile optional contract-authored query-to-physical-slot ownership.
``object_queries`` are detector prompts, not an entity roster. A contract
may therefore group several prompts as aliases of one physical subject or
attach an exact cardinality to one aggregate prompt. Every authored query
must be owned exactly once; malformed or lossy groupings block lowering.
"""
raw_groups = parameters.get("physical_subject_groups")
if raw_groups is None:
return None
if not isinstance(raw_groups, (list, tuple)) or not raw_groups:
return [], ["physical_subject_groups_missing_or_invalid"]
source_values = _dedupe(subjects)
source_set = set(source_values)
consumed: list[str] = []
specs: list[dict[str, Any]] = []
reasons: list[str] = []
for index, raw in enumerate(raw_groups):
if not isinstance(raw, Mapping):
reasons.append(f"physical_subject_group_not_object:{index}")
continue
aliases = _dedupe(
_list(raw.get("aliases"))
or [raw.get("identity_query")]
)
unknown = sorted(set(aliases) - source_set)
if not aliases:
reasons.append(f"physical_subject_group_aliases_missing:{index}")
continue
if unknown:
reasons.append(
f"physical_subject_group_unreviewed_alias:{index}:" + ",".join(unknown)
)
duplicates = sorted(set(aliases) & set(consumed))
if duplicates:
reasons.append(
f"physical_subject_query_multiply_owned:{index}:" + ",".join(duplicates)
)
try:
expected_count = int(raw.get("expected_count", 1))
except (TypeError, ValueError):
expected_count = 0
if expected_count < 1:
reasons.append(f"physical_subject_expected_count_invalid:{index}")
expected_count = 1
consumed.extend(aliases)
specs.append(
_subject_spec(
index,
aliases,
expected_count=expected_count,
subject_id=str(raw.get("id") or ""),
component_queries=_list(raw.get("component_queries")),
part_schema=_mapping(raw.get("part_schema")),
provenance="reviewed decision_rule.parameters.physical_subject_groups",
)
)
unowned = [value for value in source_values if value not in set(consumed)]
if unowned:
reasons.append("physical_subject_queries_unowned:" + ",".join(unowned))
return specs, reasons
def _inventory_subject_specs(
subjects: Sequence[str], parameters: Mapping[str, Any],
) -> tuple[list[dict[str, Any]], list[str]]:
"""Turn aggregate inventory prompts into exact physical cardinalities."""
expected_raw = parameters.get("expected_counts")
if not isinstance(expected_raw, Mapping):
return [], ["inventory_expected_counts_missing"]
expected = {
str(key): int(value)
for key, value in expected_raw.items()
if str(key) != "total"
and isinstance(value, (int, float)) and not isinstance(value, bool)
and int(value) > 0
}
values = _dedupe(subjects)
category_queries: dict[str, list[str]] = {key: [] for key in expected}
unmatched: list[str] = []
for query in values:
query_tokens = set(re.findall(r"[a-z0-9]+", query.casefold().replace("_", " ")))
ranked: list[tuple[int, str]] = []
for category in expected:
category_tokens = {
_query_token(token)
for token in re.findall(r"[a-z0-9]+", category.casefold().replace("_", " "))
}
overlap = len(category_tokens & {_query_token(token) for token in query_tokens})
if overlap:
ranked.append((overlap, category))
if ranked:
ranked.sort(key=lambda row: (-row[0], row[1]))
category_queries[ranked[0][1]].append(query)
else:
unmatched.append(query)
empty_categories = [key for key, queries in category_queries.items() if not queries]
if unmatched and len(empty_categories) == 1:
category_queries[empty_categories[0]] = unmatched
unmatched = []
specs: list[dict[str, Any]] = []
reasons: list[str] = []
for category, count in expected.items():
queries = category_queries.get(category, [])
if not queries:
reasons.append(f"inventory_category_queries_missing:{category}")
continue
if len(queries) == count:
for query in queries:
specs.append(
_subject_spec(
len(specs), [query],
provenance=(
"reviewed object_queries + "
"decision_rule.parameters.expected_counts"
),
)
)
elif len(queries) == 1:
specs.append(
_subject_spec(
len(specs), queries,
expected_count=count,
provenance=(
"reviewed aggregate object_query + "
"decision_rule.parameters.expected_counts"
),
)
)
else:
reasons.append(f"inventory_category_cardinality_ambiguous:{category}")
if unmatched:
reasons.append("inventory_queries_unassigned:" + ",".join(unmatched))
expected_total = expected_raw.get("total")
physical_total = sum(int(item["expected_count"]) for item in specs)
if isinstance(expected_total, (int, float)) and int(expected_total) != physical_total:
reasons.append(
f"inventory_total_mismatch:expected={int(expected_total)}:compiled={physical_total}"
)
return specs, reasons
def _subject_specs(
subjects: Sequence[str], source_rule_type: str, parameters: Mapping[str, Any],
) -> tuple[list[dict[str, Any]], list[str], list[str], list[str]]:
values = _dedupe(subjects)
context: list[str] = []
component_queries: list[str] = []
reasons: list[str] = []
explicit = _explicit_physical_subject_specs(values, parameters)
if explicit is not None:
specs, explicit_reasons = explicit
return specs, context, component_queries, explicit_reasons
if source_rule_type == "strict_person_instance_preservation":
specs = [
_subject_spec(
0, values,
provenance="singular reviewed person rule; object_queries are aliases",
)
] if values else []
return specs, context, component_queries, reasons
if source_rule_type == "all_instances_strict_preservation":
magazine_queries = [
value for value in values
if "magazine" in {
_query_token(token)
for token in re.findall(r"[a-z0-9]+", value.casefold())
}
]
other_queries = [value for value in values if value not in set(magazine_queries)]
specs: list[dict[str, Any]] = []
expected_parts = parameters.get("expected_magazine_count")
if magazine_queries:
part_schema = {
"kind": "stack_members",
"entity_type": "magazine",
"expected_count": int(expected_parts),
"count_provenance": "decision_rule.parameters.expected_magazine_count",
} if isinstance(expected_parts, (int, float)) else {}
specs.append(
_subject_spec(
len(specs), magazine_queries,
component_queries=magazine_queries,
part_schema=part_schema,
provenance="reviewed magazine-stack aliases",
)
)
specs.extend(
_subject_spec(len(specs) + index, [value])
for index, value in enumerate(other_queries)
)
return specs, context, component_queries, reasons
if source_rule_type == "movable_inventory_integrity":
specs, inventory_reasons = _inventory_subject_specs(values, parameters)
return specs, context, component_queries, inventory_reasons
groups: list[list[str]] = [[item] for item in values]
if source_rule_type == "relocated_object_in_air_zone":
groups, context = ([values[:2]] if values else []), values[2:]
elif source_rule_type == "remote_pair_on_table_surface":
groups, context = [[item] for item in values[:2]], values[2:]
elif source_rule_type == "paired_relocation_beside_dispenser":
groups, context = [[item] for item in values[:2]], values[2:]
elif source_rule_type == "same_unit_different_compartment_placement":
groups, context = ([values[:1]] if values else []), values[1:]
elif source_rule_type == "supported_nonintersecting_placement":
groups, context = ([values[:1]] if values else []), values[1:]
elif source_rule_type == "structure_and_tray_preservation":
# These queries name components of one registered structural system,
# not four independent physical instances.
groups = [values] if values else []
elif source_rule_type == "architecture_and_decor_preservation":
component_queries = _dedupe(_list(parameters.get("architecture_component_queries")))
decor_queries = _dedupe(_list(parameters.get("tidy_decor_queries")))
aliases = component_queries + decor_queries
specs = [
_subject_spec(
0,
aliases,
component_queries=aliases,
part_schema={
"kind": "registered_architecture_and_decor_component_roster_v1",
"major_structure_queries": component_queries,
"tidy_decor_queries": decor_queries,
"count_provenance": "reviewed decision_rule.parameters component query rosters",
},
provenance="reviewed registered architecture/decor component system",
)
] if aliases else []
return specs, context, aliases, reasons
elif source_rule_type == "upright_group_arrangement":
# The first query names the aggregate group whose ownership/location
# must be certified. Remaining queries describe components to parse
# *inside* that group; treating them as additional physical subjects
# incorrectly made one binder row plus one binder spine the expected
# roster and allowed unrelated shelf proposals into the pose metric.
groups = [values[:1]] if values else []
component_queries = values[1:]
elif source_rule_type == "masked_entity_structure_preservation":
# Detector prompts for head/torso/limbs describe components of one
# articulated entity. They are not five independent roster slots.
root = min(values, key=lambda value: (len(value.split()), len(value))) if values else ""
components = [value for value in values if value != root]
required_parts = _dedupe(parameters.get("major_parts") or [])
specs = [
_subject_spec(
0,
[root],
component_queries=components,
part_schema={
"kind": "normalized_humanoid_six_part_support_v1",
"required_parts": required_parts,
"component_queries": components,
"count_provenance": "decision_rule.parameters.major_parts",
},
provenance="reviewed articulated root query + component schema",
)
] if root else []
return specs, context, components, reasons
specs = [
_subject_spec(index, aliases)
for index, aliases in enumerate(groups) if aliases
]
return specs, context, component_queries, reasons
def lower_reviewed_predicate(check: Mapping[str, Any]) -> dict[str, Any] | None:
"""Lower one of the 29 reviewed source rules, or return ``None``."""
params = _mapping(check.get("params"))
decision_rule = _mapping(params.get("decision_rule"))
source_rule_type = str(decision_rule.get("type") or "").strip()
if source_rule_type not in RULE_FAMILY_TARGETS:
return None
candidate_value = check.get("predicate_ir_candidate") or params.get("predicate_ir_candidate")
candidate_source = reviewed_source_check_projection(check)
candidate = _mapping(candidate_value) or compile_predicate_ir_candidate(candidate_source)
candidate_validation = validate_predicate_ir_candidate(
candidate,
source_check=candidate_source,
)
# A generated candidate is cacheable only for the exact authored check.
# Regenerate it deterministically after criteria, thresholds, regions, or
# subject queries change; never let a schema-valid stale candidate force
# the new check into an old predicate program.
if "source_check_hash_mismatch" in candidate_validation.reasons:
candidate = compile_predicate_ir_candidate(candidate_source)
candidate_validation = validate_predicate_ir_candidate(
candidate,
source_check=candidate_source,
)
expected_families = list(RULE_FAMILY_TARGETS[source_rule_type])
if [str(item) for item in candidate.get("target_families", [])] != expected_families:
candidate = compile_predicate_ir_candidate(candidate_source)
candidate_validation = validate_predicate_ir_candidate(
candidate,
source_check=candidate_source,
)
# Candidate IR is a cache, never authority. A rule that gains a reviewed
# deterministic lowering must not remain manual merely because an older
# candidate recorded it as unmapped.
if not candidate_validation.valid:
regenerated = compile_predicate_ir_candidate(candidate_source)
regenerated_validation = validate_predicate_ir_candidate(
regenerated,
source_check=candidate_source,
)
if regenerated_validation.valid:
candidate = regenerated
candidate_validation = regenerated_validation
check_id = str(check.get("id") or check.get("check_id") or "").strip()
families = [str(item) for item in candidate.get("target_families", [])]
subjects = _dedupe(params.get("object_queries") or candidate.get("subjects") or [])
reasons: list[str] = []
if not candidate_validation.valid:
reasons.extend("candidate_invalid:" + item for item in candidate_validation.reasons)
if not families or any(family not in LOWERED_FAMILIES for family in families):
reasons.append("candidate_family_not_lowerable")
family = families[0] if len(families) == 1 else COMPOSITE_FAMILY
regions: list[dict[str, Any]] = []
for raw in _list(params.get("planned_regions")):
if not isinstance(raw, Mapping):
continue
region, region_reasons = _materialize_region(raw, subjects=subjects)
if (
source_rule_type == "supported_nonintersecting_placement"
and str((region.get("materializer") or {}).get("kind") or "")
== "shelf_grid_parser"
):
# Support is certified against an input/output registered grid,
# even when the authored region names only the output cell.
region["image_roles"] = ["input", "pred"]
region["source_image"] = "both"
regions.append(region)
reasons.extend(region_reasons)
source_parameters = _mapping(decision_rule.get("parameters"))
runtime_parameters = deepcopy(source_parameters)
if source_rule_type == "supported_nonintersecting_placement":
runtime_parameters.update(support_contact_geometry_profile_parameters())
if source_rule_type == "structure_and_tray_preservation":
runtime_parameters.update(registered_grid_structure_profile_parameters())
_configure_registered_grid_structure_regions(regions, subjects)
if source_rule_type == "architecture_and_decor_preservation":
runtime_parameters.update(registered_architecture_preservation_profile_parameters())
_configure_registered_architecture_regions(regions, runtime_parameters, reasons)
_configure_verified_group_roster(runtime_parameters, reasons)
if source_rule_type == "masked_entity_structure_preservation":
runtime_parameters.update(articulated_entity_preservation_profile_parameters())
root_query = min(subjects, key=lambda value: (len(value.split()), len(value)))
protected = next(
(region for region in regions if region.get("semantic_role") == "protected_region"),
None,
)
if isinstance(protected, dict):
protected["image_roles"] = ["input", "pred"]
protected["source_image"] = "both"
protected["materializer"] = {
"kind": "detector_segmenter",
"queries": [root_query],
"seed": "check_scoped_detector_boxes",
"selection_policy": "top1_whole_entity_then_sam",
"region_rule": str(protected.get("query_or_rule") or ""),
}
allowed = next(
(region for region in regions if region.get("semantic_role") == "allowed_edit_exclusion"),
None,
)
support_queries = _dedupe((allowed or {}).get("reviewed_support_queries") or [])
selector_identity = _semantic_query_head(support_queries[0]) if support_queries else ""
selector_text = " ".join(
[str((allowed or {}).get("query_or_rule") or ""), *support_queries]
).casefold()
anchor_part = (
"viewer_right_shoulder" if "viewer" in selector_text and "right" in selector_text
else "viewer_left_shoulder" if "viewer" in selector_text and "left" in selector_text
else ""
)
if isinstance(allowed, dict) and selector_identity and anchor_part:
allowed["image_roles"] = ["input"]
allowed["source_image"] = "input"
allowed["materializer"] = {
"kind": "registered_change_derivation",
"operation": "source_selector_registered_projection",
"projection_policy": "source_mask_to_registered_comparison_v1",
"source_selector_profile": "anchor_relative_source_instance_v1",
"region_rule": str(allowed.get("query_or_rule") or ""),
}
runtime_parameters["source_selector_profile"] = "anchor_relative_source_instance_v1"
runtime_parameters["source_selector"] = {
"profile": "anchor_relative_source_instance_v1",
"identity_queries": [selector_identity],
"anchor_queries": [root_query],
"anchor_part": anchor_part,
"relation": "contact_or_near",
"expected_count": 1,
"minimum_identity_score": 0.05,
"minimum_anchor_score": 0.10,
"minimum_selection_margin": 0.12,
"purpose": "allowed_edit_exclusion",
}
else:
reasons.append("articulated_allowed_edit_source_selector_not_derivable")
_configure_multi_region_protected_regions(regions, source_rule_type)
_append_parameter_allowed_edit_region(
regions,
source_parameters,
reasons,
require_complete=bool(
objective_obligation_producer(family, source_rule_type)
),
)
roster_queries = _dedupe(
str(item.get("query_text") or "")
for item in _list(source_parameters.get("source_instance_annotations"))
if isinstance(item, Mapping)
)
_scope_allowed_edit_regions(regions, subjects, roster_queries=roster_queries)
if source_rule_type != "architecture_and_decor_preservation":
_append_explicit_comparison_context_exclusion(
regions, subjects=subjects, roster_queries=roster_queries,
)
for region in regions:
materializer = region.get("materializer")
if (
str(region.get("semantic_role") or "") == "allowed_edit_exclusion"
and isinstance(materializer, Mapping)
and str(materializer.get("kind") or "") == "detector_segmenter"
and not materializer.get("queries")
):
reasons.append(
"allowed_edit_subject_queries_missing:"
+ str(region.get("region_id") or "missing")
)
required_parameters = RULE_REQUIRED_PARAMETERS.get(source_rule_type, frozenset())
missing_required_parameters = sorted(required_parameters - set(source_parameters))
if missing_required_parameters:
reasons.append("source_rule_parameter_schema_mismatch:" + ",".join(missing_required_parameters))
if "maximum_normalized_keypoint_error" in source_parameters:
pose_profile = str(source_parameters.get("pose_profile") or "")
if pose_profile != HUMAN_POSE_PROFILE:
reasons.append("normalized_keypoint_error_requires_registered_pose_profile")
local_facts_requested = any(
key in source_parameters
for key in ("max_rotation_change_degrees", "maximum_rotation_change_degrees", "minimum_local_feature_match_ratio")
)
local_profile = str(source_parameters.get("local_correspondence_profile") or "")
if local_facts_requested and local_profile and local_profile != LOCAL_CORRESPONDENCE_PROFILE:
reasons.append("local_correspondence_profile_unsupported")
bindings = compile_parameter_bindings(source_parameters)
for binding in bindings:
if binding.get("consumer") == "decision_clause":
objective_producer = objective_producer_for_binding(family, binding, runtime_parameters)
if objective_producer:
binding["producer"] = objective_producer
binding["producer_mode"] = "objective"
else:
binding["producer"] = "reviewed_semantic_vlm"
binding["producer_mode"] = "semantic_vlm"
binding["source_text"] = (
"No registered deterministic runtime producer for reviewed parameter "
+ str(binding.get("source_key") or binding.get("fact") or "")
)
if source_rule_type == "architecture_and_decor_preservation":
fail_text = str(decision_rule.get("fail_condition") or "").casefold()
broad_decor_failure = bool(
"decor" in fail_text
and any(token in fail_text for token in ("broad", "large-scale", "majority"))
)
if broad_decor_failure:
bindings.append(
{
"source_key": "derived:fail_condition:broad_tidy_decor_removal",
"consumer": "decision_clause",
"clause_id": "derived:architecture_broad_tidy_decor_removal",
"branch": "fail",
"fact": "tidy_decor_unmatched_fraction",
"op": "ge",
"expected": float(
runtime_parameters[
"architecture_broad_decor_removal_minimum_fraction"
]
),
"producer": "reviewed_typed_evidence_adapter",
"producer_mode": "objective",
"source_text": str(decision_rule.get("fail_condition") or ""),
"certificate_version": (
ARCHITECTURE_QUALITATIVE_FAILURE_CERTIFICATE_VERSION
),
"threshold_provenance": (
"shared lexical certificate: broad removal means a strict "
"majority of the reviewed registered decor mask"
),
}
)
obligation_producer = objective_obligation_producer(family, source_rule_type)
for obligation in _list(candidate.get("obligations")):
if not isinstance(obligation, Mapping):
continue
branch = str(obligation.get("branch") or "pass")
bindings.append(
{
"source_key": str(obligation.get("id") or ""),
"consumer": "decision_clause",
"clause_id": "obligation:" + str(obligation.get("id") or ""),
"branch": branch,
"fact": "obligation:" + str(obligation.get("id") or ""),
"op": "eq",
"expected": True,
"source_text": str(obligation.get("source_text") or ""),
# Free-form pass/fail prose is semantic by default. Promotion
# is allowed only through the runtime-aligned registry above,
# after a rule-specific executor proves the exact branch from
# complete objective measurements.
"producer": obligation_producer or "reviewed_semantic_vlm",
"producer_mode": "objective" if obligation_producer else "semantic_vlm",
}
)
witnesses = build_binding_witnesses(bindings)
if not witnesses or any(not bool(item.get("passed")) for item in witnesses):
reasons.append("parameter_witness_failure")
if not check_id:
reasons.append("missing_check_id")
if not subjects:
reasons.append("missing_subject_queries")
subject_specs, context_queries, component_queries, subject_reasons = _subject_specs(
subjects, source_rule_type, runtime_parameters,
)
reasons.extend(subject_reasons)
decision_bindings = [
item for item in bindings if item.get("consumer") == "decision_clause"
]
semantic_fact_requirements = [
{
"fact": str(item.get("fact") or ""),
"clause_id": str(item.get("clause_id") or ""),
"branch": str(item.get("branch") or "pass"),
"source_text": str(item.get("source_text") or ""),
"producer": str(item.get("producer") or ""),
}
for item in decision_bindings
if item.get("producer_mode") == "semantic_vlm"
]
predicate = {
"schema_version": "pvr.executable_predicate.v1",
"check_id": check_id,
"status": "executable" if not reasons else "blocked_uncompiled",
"family": family,
"target_families": families,
"source_rule_type": source_rule_type,
"source_check_sha256": candidate.get("source_check_sha256"),
"subjects": subject_specs,
"context_queries": context_queries,
"component_queries": component_queries,
"regions": regions,
"operations": _operations(families, runtime_parameters),
"state_schema": {
"unit": "typed_contract_obligation",
"states": ["pass", "partial", "fail", "unknown"],
"facts": sorted({str(item.get("fact")) for item in bindings if item.get("consumer") == "decision_clause"}),
},
"decision": {
"aggregation": "ALL_OF" if len(families) > 1 else "SINGLE_FAMILY",
"allowed_outcomes": ["pass", "partial", "fail", "unknown"],
"unknown_policy": "missing_or_ambiguous_required_fact_is_unknown",
},
"parameters": deepcopy(runtime_parameters),
"parameter_bindings": bindings,
"fact_producers": [
{
"fact": str(item.get("fact") or ""),
"clause_id": str(item.get("clause_id") or ""),
"producer": str(item.get("producer") or ""),
"mode": str(item.get("producer_mode") or ""),
}
for item in decision_bindings
],
"semantic_fact_requirements": semantic_fact_requirements,
"parameter_provenance": {
str(item.get("source_key")): "reviewed decision_rule.parameters"
for item in bindings if not str(item.get("source_key") or "").startswith(check_id + ":")
} | {
key: "derived:registered_support_contact_geometry_profile"
for key in set(runtime_parameters) - set(source_parameters)
if source_rule_type == "supported_nonintersecting_placement"
} | {
key: "derived:registered_grid_structure_preservation_profile"
for key in set(runtime_parameters) - set(source_parameters)
if source_rule_type == "structure_and_tray_preservation"
} | {
key: "derived:registered_architecture_preservation_profile"
for key in set(runtime_parameters) - set(source_parameters)
if source_rule_type == "architecture_and_decor_preservation"
},
"obligations": deepcopy(_list(candidate.get("obligations"))),
"coverage_map": {
"parameter_keys": list(source_parameters),
"region_ids": [str(item.get("region_id") or "") for item in regions],
"obligation_ids": [str(item.get("id") or "") for item in _list(candidate.get("obligations")) if isinstance(item, Mapping)],
},
"witness_results": witnesses,
"capability_versions": {
"lowerer": LOWERER_VERSION,
"decision_evaluator": "pvr.contract_obligations.v1",
"metric_binding": "pvr.contract_metric.v7",
"fact_producer_registry": "pvr.runtime_fact_producers.v1",
"semantic_fallback": "pvr.reviewed_semantic_vlm.v1",
},
"unmapped_contract_parameters": [],
"reasons": list(dict.fromkeys(reasons)),
"provenance": {
"compiler": "pvr_eval.contracts.predicate_lowering.lower_reviewed_predicate",
"candidate_schema": candidate.get("schema_version"),
"conversion": "reviewed_golden_ir_then_deterministic_lowering",
},
}
change_spec = compile_change_spec(predicate)
if change_spec:
predicate["change_spec"] = change_spec
predicate["capability_versions"]["visual_change_spec"] = "pvr.visual_change_spec.v1"
return predicate
|