"""Hesitation markers per 1k generated tokens (paragraphs starting with Wait/Alternatively/Hmm/...), per run."""
import json, re, sys
HES = re.compile(r"\n\n(\*\*)?(wait|but wait|hmm|alternatively|actually|hold on|let me (double[- ]check|verify|re-?check|reconsider|check again))", re.I)
for tag in sys.argv[1:]:
    rs = [json.loads(l) for l in open(f"results/{tag}/mini.jsonl", encoding="utf-8")]
    toks = sum(r["n_tokens"] for r in rs); h = sum(len(HES.findall(r["completion"])) for r in rs)
    print(f"{tag:12s} n={len(rs):3d}  hesitations/1k tok {1000 * h / toks:5.2f}  avg_tok {toks / len(rs):6.0f}")