File size: 4,324 Bytes
55c86fd | 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 | #!/usr/bin/env python3
"""Audit Track B story quality beyond schema validity.
This is a lightweight gate for draft assets. It does not replace human review,
but catches the common failure mode of template-filled stories that pass the
GT builder while lacking concrete visual detail.
"""
from __future__ import annotations
import argparse
import json
import re
from pathlib import Path
from statistics import mean
from typing import Any
HERE = Path(__file__).resolve().parent.parent # trackB root (scripts/ lives one level down)
GT_DIR = HERE / "gt"
SUT_DIR = HERE / "sut_prompts"
PROBE_MIN = {
"lookalike_disambiguation": 5,
"state_change": 5,
"persist_state": 5,
"count_memory": 5,
"false_friend": 5,
"deprecation_avoidance": 5,
"reference_indirect": 5,
"long_gap_reappearance": 5,
"temporal_reference": 5,
}
TEMPLATE_PATTERNS = [
r"一辆/一件",
r"一本/一卷",
r"与[^，。]{0,20}相符的",
r"开始例行检查",
r"主场日常",
r"主场中央静静",
r"主场角落坐下小憩",
r"核心器物",
r"信物",
r"载具",
r"驿馆檐下",
r"驿馆梁下",
r"驼峰",
r"月牙烙印",
r"青铜狮钮印",
r"狮鬃",
r"沙窝",
r"鼻铃",
r"重新重新",
r"过关章程",
r"认真记下当日状况",
r"关键记号",
r"一件[^，。]{0,20}细节不同",
]
def load_json(path: Path) -> dict[str, Any]:
return json.loads(path.read_text(encoding="utf-8"))
def story_ids(paths: list[str] | None) -> list[str]:
if paths:
return [Path(p).stem.replace("_name_anchored", "") for p in paths]
return [p.stem for p in sorted(GT_DIR.glob("*.json"))]
def audit_one(story_id: str) -> list[str]:
issues: list[str] = []
gt_path = GT_DIR / f"{story_id}.json"
sut_path = SUT_DIR / f"{story_id}_name_anchored.json"
if not gt_path.exists():
return [f"{story_id}: missing gt"]
gt = load_json(gt_path)
nseg = gt.get("summary", {}).get("n_segments", len(gt.get("segments", [])))
if not 50 <= nseg <= 200:
issues.append(f"{story_id}: n_segments={nseg} outside [50,200]")
probes = gt.get("summary", {}).get("probe_counts", {})
for key, minimum in PROBE_MIN.items():
if probes.get(key, 0) < minimum:
issues.append(f"{story_id}: probe {key}={probes.get(key, 0)} < {minimum}")
if gt.get("warnings"):
issues.append(f"{story_id}: gt warnings={len(gt['warnings'])}")
if sut_path.exists():
sut = load_json(sut_path)
if sut.get("warnings"):
issues.append(f"{story_id}: sut warnings={len(sut['warnings'])}")
else:
issues.append(f"{story_id}: missing sut prompt")
text = json.dumps(gt, ensure_ascii=False)
hits = sum(len(re.findall(pat, text)) for pat in TEMPLATE_PATTERNS)
if story_id not in {"0001_lighthouse_keeper", "0002_night_market_courier", "0003_desert_archaeologist"}:
if hits > 8:
issues.append(f"{story_id}: template-pattern hits={hits} > 8")
return issues
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("stories", nargs="*", help="story IDs or filenames; default: all")
parser.add_argument("--strict", action="store_true", help="exit non-zero on any issue")
args = parser.parse_args()
ids = story_ids(args.stories)
all_issues: list[str] = []
segs: list[int] = []
for sid in ids:
gt_path = GT_DIR / f"{sid}.json"
if gt_path.exists():
gt = load_json(gt_path)
segs.append(gt.get("summary", {}).get("n_segments", len(gt.get("segments", []))))
all_issues.extend(audit_one(sid))
if segs:
buckets = {
"50-80": sum(50 <= x <= 80 for x in segs),
"81-120": sum(81 <= x <= 120 for x in segs),
"121-160": sum(121 <= x <= 160 for x in segs),
"161-200": sum(161 <= x <= 200 for x in segs),
}
print(f"stories={len(ids)} segments min={min(segs)} max={max(segs)} mean={mean(segs):.1f} buckets={buckets}")
if all_issues:
print("ISSUES:")
for issue in all_issues:
print(f"- {issue}")
if args.strict:
raise SystemExit(1)
else:
print("OK")
if __name__ == "__main__":
main()
|