#!/usr/bin/env python
"""Results-to-claim audit. Reads results/claim_registry.csv (claim, metric, value, benchmark, run_ids, artifact,
status) and verifies that (i) every artifact path exists, (ii) every run id has a manifest with state completed
(or the registry explicitly says 'partial'), (iii) every numeric macro used in the manuscript exists in
paper/tables/numbers.tex, and (iv) the manuscript does not use forbidden promotional words.
Exit code 1 on any failure. Usage: python scripts/verify_manuscript_claims.py
"""
from __future__ import annotations

import csv, glob, json, re, sys
from pathlib import Path

FORBIDDEN = ["revolutionary", "state-of-the-art", "breakthrough", "groundbreaking", "unprecedented", "novel architecture that outperforms"]


def main() -> int:
    ok = True
    reg = Path("results/claim_registry.csv")
    if not reg.exists():
        print("FAIL: results/claim_registry.csv missing"); return 1
    rows = list(csv.DictReader(open(reg)))
    print(f"{len(rows)} registered claims")
    for r in rows:
        for art in filter(None, [a.strip() for a in r["artifact"].split(";")]):
            if not Path(art).exists():
                print(f"FAIL artifact missing for claim '{r['claim'][:60]}': {art}"); ok = False
        for rid in filter(None, [x.strip() for x in r["run_ids"].split(";")]):
            mans = glob.glob(f"experiments/runs/{rid}*/manifest.json") + glob.glob(f"experiments/runs/{rid}/manifest.json")
            if not mans:
                print(f"FAIL run manifest missing for '{r['claim'][:60]}': {rid}"); ok = False; continue
            st = json.load(open(mans[0])).get("state", "")
            if st != "completed" and r.get("status", "") not in ("partial", "stopped-run-used"):
                print(f"FAIL run {rid} state={st} but claim status={r.get('status')}"); ok = False
    nums = Path("paper/tables/numbers.tex").read_text() if Path("paper/tables/numbers.tex").exists() else ""
    defined = set(re.findall(r"\\newcommand\{\\([A-Za-z]+)\}", nums))
    tex = "\n".join(Path(p).read_text() for p in glob.glob("paper/sections/*.tex") + glob.glob("paper/appendix/*.tex"))
    used = set(re.findall(r"\\(params[A-Za-z]+|labels[A-Za-z]+|ctrl[A-Za-z]+|oracle[A-Za-z]+|vtwo[A-Za-z]+|latency[A-Za-z]+|cost[A-Za-z]+|ntrain[A-Za-z]+|pilot[A-Za-z]+|em[A-Z][A-Za-z]+)", tex))
    for u in sorted(used - defined):
        print(f"FAIL macro \\{u} used in manuscript but not generated"); ok = False
    low = tex.lower()
    for w in FORBIDDEN:
        if w in low:
            print(f"FAIL forbidden phrase in manuscript: '{w}'"); ok = False
    # unresolved placeholders
    for p in glob.glob("paper/sections/*.tex") + glob.glob("paper/appendix/*.tex"):
        if "% to be written" in Path(p).read_text() or "TODO" in Path(p).read_text():
            print(f"FAIL placeholder left in {p}"); ok = False
    print("AUDIT", "PASS" if ok else "FAIL")
    return 0 if ok else 1


if __name__ == "__main__":
    raise SystemExit(main())