# -*- coding: utf-8 -*-
# Re-audit every bench run log by its SCRIPT-LEVEL step lines (ground truth:
# "STEP <label> rx=<got>/<want> OK|QUIET"). probe_classify verdicts are advisory
# only; a real freeze shows a short/QUIET step here.
# ASCII output only.
import glob, os, re

def audit(path):
    steps = []
    for line in open(path, encoding="utf-8-sig", errors="replace"):
        m = re.match(r"STEP (\S+) rx=(\d+)/(\d+) (OK|QUIET)", line.strip())
        if m:
            steps.append((m.group(1), int(m.group(2)), int(m.group(3)), m.group(4)))
    bad = [s for s in steps if s[3] != "OK" or s[1] != s[2]]
    return steps, bad

groups = {}
for path in sorted(glob.glob("MDK-ARM/ab_*.log")) + sorted(glob.glob("MDK-ARM/reg_*.log")):
    name = os.path.basename(path)
    key = re.sub(r"_run\d+\.log$", "", name)
    steps, bad = audit(path)
    groups.setdefault(key, []).append((name, len(steps), bad))

for key in sorted(groups):
    runs = groups[key]
    total = len(runs)
    clean = sum(1 for _, n, bad in runs if n > 0 and not bad)
    frozen = [name for name, n, bad in runs if n > 0 and bad]
    nostep = [name for name, n, bad in runs if n == 0]
    print("%-16s runs=%2d  script-OK=%2d  script-FAIL=%2d %s" %
          (key, total, clean, len(frozen), ("FAIL:" + ",".join(x.replace(key + "_", "") for x in frozen)) if frozen else ""))
    if nostep:
        print("                 (no STEP lines: %s)" % ",".join(nostep))