import json, glob
from collections import defaultdict

files = sorted(glob.glob('/tmp/ipl_json/*.json'))
files = [f for f in files if '/README' not in f and 'people.json' not in f]

# For each match's 2nd innings, walk ball-by-ball and record (balls_remaining_bucket, runs_required_bucket, wickets_in_hand) -> did chasing team win
# Buckets: balls_remaining in [0,120] step 6 (over-level); runs_required capped/bucketed; wickets_in_hand 0-10
outcomes = defaultdict(lambda: [0, 0])  # key -> [wins, total]
matches_processed = 0
skipped = 0
per_over_totals = defaultdict(list)  # over_number(0-19) -> list of runs scored that over (for "par score" reference)

for f in files:
    try:
        d = json.load(open(f))
    except Exception:
        skipped += 1; continue
    info = d['info']
    if info.get('match_type') != 'T20': 
        pass  # IPL is all T20, keep anyway
    outcome = info.get('outcome', {})
    winner = outcome.get('winner')
    if not winner or 'result' in outcome:  # skip no-result/tied edge cases for simplicity
        if outcome.get('result') in ('tie','no result'):
            skipped += 1; continue
    innings = d.get('innings', [])
    if len(innings) < 2:
        skipped += 1; continue

    inn1 = innings[0]
    target = 0
    for ov in inn1['overs']:
        for ball in ov['deliveries']:
            target += ball['runs']['total']
            # first-innings per-over tracking (for par score reference)
    # recompute target properly and per-over totals for innings 1 (par score reference)
    target = 0
    over_runs = defaultdict(int)
    for ov in inn1['overs']:
        for ball in ov['deliveries']:
            target += ball['runs']['total']
            over_runs[ov['over']] += ball['runs']['total']
    for ov_num, runs in over_runs.items():
        per_over_totals[ov_num].append(runs)
    target += 1  # runs needed to win

    inn2 = innings[1]
    chasing_team = inn2['team']
    chasing_won = (winner == chasing_team)

    balls_bowled = 0
    runs_scored = 0
    wickets_down = 0
    TOTAL_BALLS = 120  # T20 = 20 overs * 6, ignore rare non-6-ball overs for bucketing simplicity

    for ov in inn2['overs']:
        for ball in ov['deliveries']:
            balls_bowled += 1
            runs_scored += ball['runs']['total']
            is_wicket = 'wickets' in ball
            # snapshot state BEFORE this ball resolves (i.e., using pre-ball counts) then update
            balls_remaining = max(0, TOTAL_BALLS - balls_bowled + 1)  # state at start of this ball
            runs_required = max(0, target - (runs_scored - ball['runs']['total']))
            wkts_in_hand = 10 - wickets_down

            # bucket: balls_remaining to nearest 6 (over granularity), runs_required to nearest 5, wkts_in_hand exact
            br_bucket = round(balls_remaining/6)*6
            rr_bucket = round(runs_required/5)*5
            key = (br_bucket, rr_bucket, wkts_in_hand)
            outcomes[key][1] += 1
            if chasing_won:
                outcomes[key][0] += 1

            if is_wicket:
                wickets_down += len(ball['wickets'])
    matches_processed += 1

print(f"Processed {matches_processed} matches, skipped {skipped}")
print(f"Unique game states: {len(outcomes)}")

# Save the win-probability lookup table (only keep states with enough samples for reliability)
reliable = {f"{k[0]},{k[1]},{k[2]}": {"wins": v[0], "total": v[1], "win_pct": round(100*v[0]/v[1],1)} for k,v in outcomes.items() if v[1] >= 15}
print(f"Reliable states (n>=15): {len(reliable)}")

json.dump({
    'meta': {
        'source': 'Cricsheet ball-by-ball IPL data (cricsheet.org)',
        'matches_processed': matches_processed,
        'method': 'For every ball of every 2nd innings in 1,243 IPL matches (2008-2026), record game state (balls remaining, runs required, wickets in hand) and whether the chasing team ultimately won. Bucketed to over-level balls-remaining and nearest-5 runs-required for sample density. States with fewer than 15 historical observations excluded as unreliable.',
        'total_states': len(outcomes),
        'reliable_states': len(reliable),
    },
    'win_prob_table': reliable,
}, open('/tmp/winprob_table.json','w'))
print("saved")