File size: 9,295 Bytes
37cb069 470bcea 37cb069 470bcea 37cb069 470bcea 37cb069 470bcea 37cb069 470bcea 37cb069 470bcea 37cb069 | 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 | """Panel-quality snapshot.
Reads every reviews/<id>_*.md and emits a JSON snapshot the website can
render at /stats. Durability > volume: the dashboard is insurance against
the panel silently rubber-stamping as traffic scales.
Metrics:
- Total reviewed (all time) and within rolling 30-day window
- Recommendation mix (RECOMMEND / REVIEW_FURTHER / REVISE_AND_RESUBMIT / REJECT / PAUSED_AI_FAILURE)
- Disagreement rate (fraction where reviewers split verdicts)
- Per-dimension mean-of-means distribution (histogram bins)
- AI provenance-flag rate (fraction where ai_provenance_signal mean ≤ 2)
No model/vendor identities leak — this snapshot is safe to publish.
"""
from __future__ import annotations
import datetime as _dt
import json
import os
import re
from collections import Counter
RECOMMENDATIONS = ("RECOMMEND", "REVIEW_FURTHER", "REVISE_AND_RESUBMIT", "REJECT", "PAUSED_AI_FAILURE")
DIMENSIONS = (
"Domain Fit",
"Methodological Transparency",
"Internal Consistency",
"Citation Integrity",
"Novelty Signal",
"AI Provenance Signal",
)
def _parse_frontmatter(text: str) -> dict:
if not text.startswith("---\n"):
return {}
end = text.find("\n---\n", 4)
if end < 0:
return {}
out: dict = {}
for line in text[4:end].splitlines():
if ":" not in line:
continue
k, v = line.split(":", 1)
out[k.strip()] = v.strip().strip('"').strip("'")
return out
def _parse_aggregate_means(text: str) -> dict[str, float]:
"""Pull dimension → mean from the aggregate markdown table."""
means: dict[str, float] = {}
in_table = False
for line in text.splitlines():
stripped = line.strip()
if stripped.startswith("## Aggregate Scores"):
in_table = True
continue
if in_table and stripped.startswith("## "):
break
if not in_table or not stripped.startswith("|"):
continue
cells = [c.strip() for c in stripped.strip("|").split("|")]
if len(cells) < 2 or cells[0].lower() == "dimension":
continue
if set("".join(cells)) <= set("- "):
continue
label = cells[0]
try:
means[label] = float(cells[1])
except ValueError:
continue
return means
def _parse_review_date(raw: str) -> _dt.datetime | None:
try:
return _dt.datetime.fromisoformat(raw.replace("Z", "+00:00"))
except Exception:
return None
def _load_rqc_flags(reviews_dir: str) -> dict[str, bool]:
"""Parse review_quality_control_flag from every RQC file by record_id.
Returns {record_id: bool}. Missing RQC file for a record yields no key;
the flag-rate metric excludes un-audited records from both numerator
and denominator.
"""
flags: dict[str, bool] = {}
if not os.path.isdir(reviews_dir):
return flags
for name in sorted(os.listdir(reviews_dir)):
if not name.endswith("_review_quality_control.md"):
continue
path = os.path.join(reviews_dir, name)
with open(path, "r", encoding="utf-8") as f:
fm = _parse_frontmatter(f.read())
rid = str(fm.get("record_id", name.split("_", 1)[0]))
raw = str(fm.get("review_quality_control_flag", "false")).lower()
flags[rid] = raw == "true"
return flags
def _load_reviews(reviews_dir: str) -> list[dict]:
out: list[dict] = []
if not os.path.isdir(reviews_dir):
return out
# Defense-in-depth guard: stats are computed off the per-paper review
# markdown files in reviews/, NOT off the audit log. Test submissions
# bail at the intake handler before the panel runs and therefore
# never produce <id>_*.md files in this directory. The guard below
# asserts the caller did not accidentally hand us a glob that pulls
# in audit-log-test.jsonl alongside the markdown set; if anyone ever
# rewires _load_reviews, this trips before contamination can happen.
assert "test" not in os.path.basename(reviews_dir.rstrip("/")), (
f"stats.py refuses to read from a directory whose basename "
f"contains 'test': {reviews_dir!r}"
)
rqc_flags = _load_rqc_flags(reviews_dir)
for name in sorted(os.listdir(reviews_dir)):
if not name.endswith(".md"):
continue
if name.endswith("_review_quality_control.md"):
# Folded in via rqc_flags; not a panel review.
continue
if name.endswith("_citations.md"):
# Pre-review citation verification artifact, not a panel review.
continue
if "ICSAC-SUB-TEST-" in name:
# Belt-and-suspenders: if a test review file ever does end up
# in reviews/ (e.g. from a hand-run experiment), skip it so
# public stats never count test data.
continue
path = os.path.join(reviews_dir, name)
with open(path, "r", encoding="utf-8") as f:
text = f.read()
fm = _parse_frontmatter(text)
means = _parse_aggregate_means(text)
rid = str(fm.get("record_id", name.split("_", 1)[0]))
out.append(
{
"record_id": rid,
"recommendation": fm.get("recommendation", "REVIEW_FURTHER"),
"disagreement": fm.get("disagreement", "False").lower() == "true",
"review_date": _parse_review_date(fm.get("review_date", "")),
"dimension_means": means,
"rqc_flag": rqc_flags.get(rid),
}
)
return out
def _histogram(values: list[float]) -> dict[str, int]:
"""Distribute 1.0–5.0 scores into five 1-wide bins."""
bins = {"1-1.99": 0, "2-2.99": 0, "3-3.99": 0, "4-4.99": 0, "5": 0}
for v in values:
if v >= 5:
bins["5"] += 1
elif v >= 4:
bins["4-4.99"] += 1
elif v >= 3:
bins["3-3.99"] += 1
elif v >= 2:
bins["2-2.99"] += 1
else:
bins["1-1.99"] += 1
return bins
def compute_stats(reviews_dir: str) -> dict:
reviews = _load_reviews(reviews_dir)
now = _dt.datetime.now(_dt.timezone.utc)
cutoff = now - _dt.timedelta(days=30)
window = [r for r in reviews if r["review_date"] and r["review_date"] >= cutoff]
rec_counts = Counter(r["recommendation"] for r in reviews)
rec_counts_30d = Counter(r["recommendation"] for r in window)
disagree_30d = sum(1 for r in window if r["disagreement"])
dim_hist: dict[str, dict[str, int]] = {}
dim_means: dict[str, float] = {}
for dim in DIMENSIONS:
vals = [r["dimension_means"][dim] for r in reviews if dim in r["dimension_means"]]
dim_hist[dim] = _histogram(vals)
dim_means[dim] = round(sum(vals) / len(vals), 2) if vals else 0.0
provenance_hits = sum(
1 for r in reviews if r["dimension_means"].get("AI Provenance Signal", 5) <= 2
)
total = len(reviews)
total_30d = len(window)
# RQC flag-rate: only count records that were actually audited.
# A None rqc_flag means RQC did not run (older reviews pre-rollout).
audited = [r for r in reviews if r.get("rqc_flag") is not None]
audited_30d = [r for r in window if r.get("rqc_flag") is not None]
rqc_flagged = sum(1 for r in audited if r["rqc_flag"])
rqc_flagged_30d = sum(1 for r in audited_30d if r["rqc_flag"])
def _rate(num: int, denom: int) -> float:
return round(num / denom, 3) if denom else 0.0
return {
"generated_at": now.strftime("%Y-%m-%dT%H:%M:%SZ"),
"total_reviewed": total,
"total_reviewed_30d": total_30d,
"recommendation_mix": {r: rec_counts.get(r, 0) for r in RECOMMENDATIONS},
"recommendation_mix_30d": {r: rec_counts_30d.get(r, 0) for r in RECOMMENDATIONS},
"reject_rate_30d": _rate(rec_counts_30d.get("REJECT", 0), total_30d),
"recommend_rate_30d": _rate(rec_counts_30d.get("RECOMMEND", 0), total_30d),
"disagreement_rate_30d": _rate(disagree_30d, total_30d),
"provenance_hit_rate_overall": _rate(provenance_hits, total),
"dimension_means_overall": dim_means,
"dimension_distribution_overall": dim_hist,
"rqc_audited_count": len(audited),
"rqc_audited_count_30d": len(audited_30d),
"rqc_flagged_count_30d": rqc_flagged_30d,
"rqc_flag_rate_overall": _rate(rqc_flagged, len(audited)),
"rqc_flag_rate_30d": _rate(rqc_flagged_30d, len(audited_30d)),
}
def write_stats(reviews_dir: str, out_path: str) -> str:
stats = compute_stats(reviews_dir)
os.makedirs(os.path.dirname(out_path), exist_ok=True)
with open(out_path, "w", encoding="utf-8") as f:
json.dump(stats, f, indent=2, ensure_ascii=False)
f.write("\n")
return out_path
if __name__ == "__main__":
import sys
rdir = (
sys.argv[1]
if len(sys.argv) > 1
else os.path.join(os.path.dirname(os.path.abspath(__file__)), "reviews")
)
out = (
sys.argv[2]
if len(sys.argv) > 2
else os.path.expanduser(
"~/Desktop/icsac/icsacinstitute.org/src/data/stats.json"
)
)
written = write_stats(rdir, out)
print(f"wrote {written}")
|