triage-failures

name	triage-failures
description	Use when user wants to Analyze OCR failure corpus to find systematic misreads, top error patterns, and whether unlisted items need rules added
disable-model-invocation	true

Analyze the OCR failure corpus to surface systematic misread patterns:

uv run python - <<'EOF'
import orjson, collections
from pathlib import Path

REPO = Path(".")
files = {
 "captured": REPO / "artifacts/ocr/skip_unlisted/samples.jsonl",
 "fixed": REPO / "artifacts/ocr/failure_corpus.jsonl",
}

for label, path in files.items():
 if not path.exists():
 print(f"[{label}] not found: {path}")
 continue
 samples = [orjson.loads(l) for l in path.read_text().splitlines() if l.strip()]
 print(f"\n=== {label} ({len(samples)} samples) ===")
 if not samples:

 # Top raw→chosen mismatches
 mismatches = [(s.get("raw_text",""), s.get("chosen_name","")) for s in samples if s.get("raw_text") != s.get("chosen_name")]
 counter = collections.Counter(mismatches).most_common(10)
 print("Top raw→chosen mismatches:")
 for (raw, chosen), count in counter:
 print(f" {count:3d}x {raw!r:40s} → {chosen!r}")

 # Items with no match (matched_name is null)
 no_match = [s.get("chosen_name","") for s in samples if not s.get("matched_name")]
 if no_match:
 top_no_match = collections.Counter(no_match).most_common(10)
 print("Items with no rule match (consider adding rules):")
 for name, count in top_no_match:
 print(f" {count:3d}x {name!r}")

 # Source breakdown
 sources = collections.Counter(s.get("source","?") for s in samples)
 print(f"Source: {dict(sources)}")
EOF

Interpreting the output:

• Same raw_text maps to wrong chosen_name: OCR preprocessing issue - check binarization or upscale
• matched_name is null repeatedly: Item has no rule - use /add-rule skill
• Infobox vs context_menu skew: One detection path is degraded - run /dry-run to compare
• High raw≠chosen count: Fuzzy threshold may be too low - check score_cutoff in item_actions.py

After identifying patterns, run /corpus-replay to validate any threshold or preprocessing changes against the fixed corpus.