"""Aggregate multi-seed coloring results -> mean+/-std per architecture/config."""
import glob, json
import numpy as np
from collections import defaultdict

R = '/home/yurenh2/rrog/runs'


def ms(xs):
    a = np.array([x for x in xs if x is not None], dtype=float)
    return f"{a.mean():.3f}±{a.std():.3f} (n={len(a)})" if len(a) else "—"


def load(pat):
    out = []
    for f in glob.glob(pat):
        try:
            out.append(json.load(open(f)))
        except Exception:
            pass
    return out


def key(d):
    return (d.get('arch', 'legacy'), d.get('conv', 'gin'), d.get('pe'),
            d.get('grad_mode'), 'ctr' if d.get('contract') else '-')


solve, le, ml = defaultdict(list), defaultdict(list), defaultdict(list)
pk, ls, au, det = defaultdict(list), defaultdict(list), defaultdict(list), defaultdict(list)

for d in load(f"{R}/color_*.json"):
    if 'solve_rate' in d and d.get('pe') is not None:
        solve[key(d)].append(d['solve_rate'])
for d in load(f"{R}/le_color_*.json"):
    le[key(d)].append(d.get('auroc'))
    ml[key(d)].append(d.get('mean_lam'))
for d in load(f"{R}/ptrm_color_*.json"):
    k = key(d); det[k].append(d.get('det'))
    s2 = d.get('sigmas', {}).get('0.2')
    if s2:
        pk[k].append(s2.get('passk')); ls[k].append(s2.get('lamsel')); au[k].append(s2.get('auroc'))

print("=== best solve_rate (deterministic, EMA) ===")
for k in sorted(solve, key=str):
    print(f"  {k}: {ms(solve[k])}")
print("=== LE AUROC(fail|lambda1) ===")
for k in sorted(le, key=str):
    print(f"  {k}: {ms(le[k])}")
print("=== LE mean_lambda1 (forced-contraction dose) ===")
for k in sorted(ml, key=str):
    print(f"  {k}: {ms(ml[k])}")
print("=== PTRM sigma=0.2 ===")
for k in sorted(pk, key=str):
    print(f"  {k}: det {ms(det[k])} | pass@K {ms(pk[k])} | lambda-sel {ms(ls[k])} | AUROC {ms(au[k])}")