# Random-baseline control for the v3/v4/v4b 150-step ruler (ar25+bp35).
# Linear code only (exec globals/locals separation — HANDOVER pitfall 7).
# Uniform-random policy over available_actions, RESET only on GAME_OVER
# (mirrors run_game lifecycle). Seed fixed per game index for reproducibility.
import os, sys, json, time, random

os.chdir('/content/qwenjev')
sys.path.insert(0, '/content/qwenjev')
sys.path.insert(0, '/content/qwenjev/train')

from data.collect_selfplay import make_env, meta_from
from data.mock_env import step_env

GAMES = [('bp35', 150)]
BASE_SEED = 42

report = []
t0 = time.time()
for idx, (game_id, max_actions) in enumerate(GAMES):
    rng = random.Random(BASE_SEED + idx * 100)
    env = make_env(game_id, None)
    env.reset()
    actions = 0
    resets = 0
    best_level = 0
    final_state = 'NOT_FINISHED'
    started = time.time()
    action_hist = {}
    for _ in range(max_actions):
        meta = meta_from(env)
        final_state = meta['state']
        if final_state == 'WIN':
            break
        if final_state == 'GAME_OVER':
            env.reset()
            resets += 1
            continue
        best_level = max(best_level, meta['levels_completed'])
        avail = list(meta['available_actions']) or ['ACTION1']
        a = rng.choice([x for x in avail if x != 'RESET'] or ['ACTION1'])
        action_hist[a] = action_hist.get(a, 0) + 1
        step_env(env, a)
        actions += 1
        best_level = max(best_level, meta_from(env)['levels_completed'])
    rec = {
        'game_id': game_id,
        'win': final_state == 'WIN',
        'final_state': final_state,
        'best_level': best_level,
        'resets': resets,
        'actions': actions,
        'seconds': round(time.time() - started, 1),
        'action_hist': action_hist,
    }
    report.append(rec)
    print('RESULT', json.dumps(rec), flush=True)

out = {
    'meta': {
        'backend': 'random-uniform', 'seed_note_v2': 'budget paired with v4b-600 ar25 probe',
        'seed_note': 'rng = Random(42 + idx*100) per run, fixed for reproducibility',
        'games': [f'{g}@{m}' for g, m in GAMES],
        'max_actions': [m for _, m in GAMES],
        'seconds_total': round(time.time() - t0, 1),
        'timestamp': time.strftime('%F %T'),
    },
    'games': report,
    'summary': {
        'played': len(report),
        'wins': sum(1 for r in report if r['win']),
        'best_level_sum': sum(r['best_level'] for r in report),
        'resets_sum': sum(r['resets'] for r in report),
        'actions_sum': sum(r['actions'] for r in report),
    },
}
open('/content/eval_random_baseline.json', 'w').write(json.dumps(out, indent=2))
print('SUMMARY', json.dumps(out['summary']), flush=True)