File size: 10,475 Bytes
7aceaa5 | 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 | #!/usr/bin/env python3
from __future__ import annotations
import argparse
import csv
import json
from pathlib import Path
from typing import Iterable
ROOT = Path(__file__).resolve().parents[1]
RULE_CARDS_PATH = ROOT / "data" / "rules" / "rule_cards_starter.csv"
A_TRACES_PATH = ROOT / "data" / "canonical" / "A_canonical_traces_v1.jsonl"
B_TRACES_PATH = ROOT / "data" / "counterfactuals" / "B_counterfactual_traces_v1.jsonl"
BUNDLES_PATH = ROOT / "data" / "bundles" / "pilot_bundles_v1.jsonl"
ALLOWED_ACTION_TYPES = {
"ask_profile",
"confirm_preference",
"assess",
"recommend",
"redirect",
"disclose_risk",
"disclose_conflict",
"marketing_claim",
}
REQUIRED_BUNDLE_LABELS = {"compliant", "deadline_missed", "hard_violation"}
DEADLINE_VIOLATION_ACTION_TYPES = {"recommend", "marketing_claim"}
REMEDIAL_ACTION_TYPES = {
"ask_profile",
"confirm_preference",
"disclose_risk",
"disclose_conflict",
}
def load_jsonl(path: Path) -> list[dict]:
with path.open("r", encoding="utf-8") as handle:
return [json.loads(line) for line in handle if line.strip()]
def load_json(path: Path) -> dict:
with path.open("r", encoding="utf-8") as handle:
return json.load(handle)
def load_rule_ids(path: Path) -> set[str]:
with path.open("r", encoding="utf-8", newline="") as handle:
return {row["rule_id"] for row in csv.DictReader(handle)}
def earliest_hard_violation(trace: dict) -> int | None:
for step in trace["steps"]:
if step["hard_violation"]:
return step["step_id"]
return None
def has_hard_violation(trace: dict) -> bool:
return earliest_hard_violation(trace) is not None
def validate_trace(trace: dict, valid_rule_ids: set[str], errors: list[str], origin: str) -> None:
expected_first = earliest_hard_violation(trace)
if trace["first_violation_step"] != expected_first:
errors.append(
f"{origin}: first_violation_step={trace['first_violation_step']} but earliest hard violation is {expected_first}"
)
expected_compliant = expected_first is None and trace["label"] == "compliant"
if trace["overall_compliant"] != expected_compliant:
errors.append(
f"{origin}: overall_compliant={trace['overall_compliant']} but expected {expected_compliant}"
)
cumulative: dict[str, float] = {}
for step in trace["steps"]:
if step["action_type"] not in ALLOWED_ACTION_TYPES:
errors.append(f"{origin}: unsupported action_type {step['action_type']}")
for rule_id in step["active_rule_ids"]:
if rule_id not in valid_rule_ids:
errors.append(f"{origin}: unknown active_rule_id {rule_id}")
violated_rule_id = step["violated_rule_id"]
if violated_rule_id is not None and violated_rule_id not in valid_rule_ids:
errors.append(f"{origin}: unknown violated_rule_id {violated_rule_id}")
for rule_id, delta in step["soft_coverage_delta"].items():
if rule_id not in valid_rule_ids:
errors.append(f"{origin}: unknown soft_coverage_delta rule_id {rule_id}")
cumulative[rule_id] = cumulative.get(rule_id, 0.0) + float(delta)
if cumulative[rule_id] > 1.000001:
errors.append(
f"{origin}: cumulative soft coverage for {rule_id} exceeds 1.0 ({cumulative[rule_id]:.3f})"
)
def validate_canonical_trace_length(trace: dict, errors: list[str], origin: str) -> None:
step_count = len(trace["steps"])
if not 3 <= step_count <= 5:
errors.append(f"{origin}: canonical trace has {step_count} steps; expected 3 to 5")
def validate_deadline_missed_temporal(trace: dict, warnings: list[str], origin: str) -> None:
if trace["label"] != "deadline_missed":
return
first_violation_step = trace["first_violation_step"]
if first_violation_step is None:
warnings.append(f"{origin}: deadline_missed trace has null first_violation_step")
return
if first_violation_step > 2:
warnings.append(
f"{origin}: deadline_missed first_violation_step={first_violation_step}; expected early violation at step 1 or 2"
)
step_lookup = {step["step_id"]: step for step in trace["steps"]}
violating_step = step_lookup.get(first_violation_step)
if violating_step is None:
warnings.append(
f"{origin}: first_violation_step={first_violation_step} does not exist in steps"
)
return
violating_action = violating_step["action_type"]
if violating_action not in DEADLINE_VIOLATION_ACTION_TYPES:
warnings.append(
f"{origin}: deadline_missed first violation occurs on action_type={violating_action}; expected recommend or marketing_claim"
)
for step in trace["steps"]:
if step["step_id"] >= first_violation_step:
continue
if step["action_type"] in REMEDIAL_ACTION_TYPES:
warnings.append(
f"{origin}: remedial action_type={step['action_type']} appears before first_violation_step={first_violation_step}"
)
break
def validate_bundle(bundle: dict, errors: list[str], warnings: list[str]) -> None:
bundle_id = bundle["bundle_id"]
candidates = bundle["candidates"]
label_set = {candidate["label"] for candidate in candidates}
missing_labels = sorted(REQUIRED_BUNDLE_LABELS - label_set)
if missing_labels:
errors.append(f"{bundle_id}: missing required labels {missing_labels}")
for candidate in candidates:
label = candidate["label"]
candidate_origin = f"{bundle_id}/{candidate['trace_id']}"
candidate_has_hard_violation = has_hard_violation(candidate)
first_violation_step = candidate["first_violation_step"]
if label == "compliant" and (
candidate_has_hard_violation or first_violation_step is not None
):
warnings.append(
f"{candidate_origin}: compliant candidate has hard_violation={candidate_has_hard_violation} and first_violation_step={first_violation_step}"
)
elif label == "hard_violation" and (
not candidate_has_hard_violation or first_violation_step is None
):
warnings.append(
f"{candidate_origin}: hard_violation candidate has hard_violation={candidate_has_hard_violation} and first_violation_step={first_violation_step}"
)
elif label == "deadline_missed" and not candidate_has_hard_violation:
warnings.append(f"{candidate_origin}: deadline_missed candidate has no hard violation")
first_positions: dict[str, int] = {}
for index, candidate in enumerate(candidates):
first_positions.setdefault(candidate["label"], index)
if all(label in first_positions for label in REQUIRED_BUNDLE_LABELS):
ordered_positions = [
first_positions["compliant"],
first_positions["deadline_missed"],
first_positions["hard_violation"],
]
if ordered_positions != sorted(ordered_positions):
warnings.append(
f"{bundle_id}: candidate order violates severity ordering compliant < deadline_missed < hard_violation; positions={ordered_positions}"
)
def build_result(errors: list[str], warnings: list[str], traces_count: int, bundles_count: int) -> dict:
return {
"ok": len(errors) == 0,
"errors": errors,
"warnings": warnings,
"validated_traces": traces_count,
"validated_bundles": bundles_count,
}
def validate_dataset(a_traces: list[dict], b_traces: list[dict], bundles: list[dict], valid_rule_ids: set[str]) -> dict:
traces = a_traces + b_traces
errors: list[str] = []
warnings: list[str] = []
for trace in traces:
validate_trace(trace, valid_rule_ids, errors, trace["trace_id"])
validate_deadline_missed_temporal(trace, warnings, trace["trace_id"])
for trace in a_traces:
validate_canonical_trace_length(trace, errors, trace["trace_id"])
for bundle in bundles:
for candidate in bundle["candidates"]:
validate_trace(candidate, valid_rule_ids, errors, candidate["trace_id"])
validate_bundle(bundle, errors, warnings)
return build_result(errors, warnings, len(traces), len(bundles))
def validate_single_bundle(bundle: dict, valid_rule_ids: set[str]) -> dict:
errors: list[str] = []
warnings: list[str] = []
for candidate in bundle["candidates"]:
validate_trace(candidate, valid_rule_ids, errors, candidate["trace_id"])
validate_deadline_missed_temporal(candidate, warnings, candidate["trace_id"])
if candidate["trace_id"].endswith("_A"):
validate_canonical_trace_length(candidate, errors, candidate["trace_id"])
validate_bundle(bundle, errors, warnings)
return build_result(errors, warnings, len(bundle["candidates"]), 1)
def load_annotation_bundles(annotations_dir: Path) -> list[dict]:
bundles: list[dict] = []
for path in sorted(annotations_dir.rglob("*.json")):
bundles.append(load_json(path))
return bundles
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Validate CPRM pilot traces, bundles, or saved annotations.")
parser.add_argument(
"--bundle-json",
type=Path,
help="Validate a single saved bundle JSON file instead of the default dataset.",
)
parser.add_argument(
"--annotations-dir",
type=Path,
help="Validate all JSON bundle files under an annotations directory instead of the default dataset.",
)
return parser.parse_args()
def main() -> None:
args = parse_args()
valid_rule_ids = load_rule_ids(RULE_CARDS_PATH)
if args.bundle_json:
result = validate_single_bundle(load_json(args.bundle_json), valid_rule_ids)
elif args.annotations_dir:
bundles = load_annotation_bundles(args.annotations_dir)
result = validate_dataset([], [], bundles, valid_rule_ids)
else:
result = validate_dataset(
load_jsonl(A_TRACES_PATH),
load_jsonl(B_TRACES_PATH),
load_jsonl(BUNDLES_PATH),
valid_rule_ids,
)
print(json.dumps(result, indent=2))
if not result["ok"]:
raise SystemExit(1)
if __name__ == "__main__":
main()
|