| """Garbage (page chrome) assessment against the SOT. |
| |
| Unit level: every A0 unit is SOT-garbage when >= 80% of its non-whitespace |
| characters lie in SOT garbage segments or in text no SOT content segment |
| covers; otherwise SOT-content. Detectors: A0's own artifact_candidates |
| (possible_page_number, repeated_short_text, either) and a chrome regex |
| (page numbers / "Page n", signature-page and exhibit footers, confidentiality |
| legends, file names). Also: what the SOT's garbage segments are, by kind. |
| Usage: uv run python claude_run/NEW/scripts/42_garbage.py [run_dir] |
| """ |
| import json |
| import re |
| import sys |
| from collections import Counter, defaultdict |
| from pathlib import Path |
|
|
| ROOT = Path(__file__).resolve().parents[1] |
| RUN = Path(sys.argv[1]) if len(sys.argv) > 1 else ROOT / "ws_sot/runs/a0" |
| CHROME = re.compile(r"^\W*(page\s*)?[\divxlc]{1,4}(\s*of\s*\d+)?\W*$|^\W*[A-Z]{0,3}-?\s?\d{1,3}\W*$|signature page|\[?\s*remainder of (this )?page|" |
| r"confidential|omitted and filed separately|\.htm\b|^\W*exhibit\s+[\w.()-]+\W*$|^\W*\*+\W*$", re.I) |
| KINDS = [("page number", re.compile(r"^\W*(page\s*)?[\divxlc]{1,4}(\s*of\s*\d+)?\W*$|^\W*[A-Z]{0,3}-\s?\d{1,3}\W*$", re.I)), |
| ("signature-page footer", re.compile(r"signature page|remainder of (this )?page", re.I)), |
| ("confidentiality legend", re.compile(r"confidential|omitted and filed", re.I)), |
| ("exhibit/filing header", re.compile(r"\.htm|^\W*ex(hibit)?[\s-]*[\d.()a-z-]+\W*$", re.I)), |
| ("empty / symbols", re.compile(r"^\W*$"))] |
|
|
|
|
| def prf(tp, fp, fn): |
| p = tp / (tp + fp) if tp + fp else 0.0; r = tp / (tp + fn) if tp + fn else 0.0 |
| return p, r, (2 * p * r / (p + r) if p + r else 0.0) |
|
|
|
|
| content = defaultdict(list); gkind = Counter(); ex = defaultdict(list) |
| for line in (ROOT / "sot_provisioned.jsonl").open(): |
| r = json.loads(line); x = r["source"]["ranges"][0] |
| if not r["is_garbage"]: content[r["document_id"]].append((x["start"], x["end"])); continue |
| t = " ".join(r["literal"].split()); k = next((n for n, rx in KINDS if rx.search(t)), "other") |
| gkind[k] += 1 |
| if len(ex[k]) < 4: ex[k].append(t[:70]) |
| tot = sum(gkind.values()) |
| print(f"SOT garbage segments placed: {tot}") |
| for k, v in gkind.most_common(): print(f" {v:5} ({100 * v / tot:4.1f}%) {k}: {ex[k]}") |
| cnt = defaultdict(Counter) |
| for line in (RUN / "projection.jsonl").open(): |
| r = json.loads(line); r = r.get("current", r) |
| if r["record_type"] != "segment": continue |
| did = r["source"]["document_id"]; lit = r["literal"]; a = r["source"]["ranges"][0]["start"] |
| ns = [a + i for i, c in enumerate(lit) if not c.isspace()] |
| if not ns: continue |
| inside = sum(1 for p in ns if any(s <= p < e for s, e in content[did])) |
| truth = inside <= 0.2 * len(ns) |
| kinds = {c["kind"] for c in r["projection"]["artifact_candidates"]} |
| det = {"A0 possible_page_number": "possible_page_number" in kinds, "A0 repeated_short_text": "repeated_short_text" in kinds, |
| "A0 either flag": bool(kinds), "chrome regex": bool(CHROME.search(" ".join(lit.split()))), |
| "chrome regex OR page-number flag": bool(CHROME.search(" ".join(lit.split()))) or "possible_page_number" in kinds} |
| for name, d in det.items(): cnt[name]["tp" if d and truth else "fp" if d else "fn" if truth else "tn"] += 1 |
| print(f"\nA0 units judged SOT-garbage: {cnt['A0 either flag']['tp'] + cnt['A0 either flag']['fn']}") |
| print(f"{'detector':36} {'P':>6} {'R':>6} {'F1':>6} tp/fp/fn") |
| for name, c in cnt.items(): |
| p, r_, f = prf(c["tp"], c["fp"], c["fn"]); print(f"{name:36} {p:6.3f} {r_:6.3f} {f:6.3f} {c['tp']}/{c['fp']}/{c['fn']}") |
|
|
