"""Fixed, transparent exploratory monthly screen; no tuned or predictive rating."""
import math
import statistics as st
import random

VERSION = 'monthly-candidates-1.0'
WEIGHTS = {'trend': .30, 'relative_strength': .20, 'spot_buying': .20, 'risk': .15, 'liquidity': .15}
DAY = 86400


def clip(x):
    return min(1., max(0., x))


def valid_bar(r):
    return all(math.isfinite(r[k]) for k in ('o', 'h', 'l', 'c', 'qv')) and (
        0 < r['l'] <= min(r['o'], r['c']) <= max(r['o'], r['c']) <= r['h'] and r['qv'] > 0)


def window(series, end, n):
    rows = [series.get(end - i * DAY) for i in reversed(range(n))]
    return rows if all(r is not None and valid_bar(r) for r in rows) else None


def score(symbol, series, btc, end, *, min_median_quote=5e6):
    rows, benchmark = window(series, end, 91), window(btc, end, 91)
    if not rows or not benchmark:
        return None, 'missing_or_invalid_91_day_price_window'
    recent = rows[-30:]
    if any(r.get('taker_buy_valid') != 1 or not math.isfinite(r.get('taker_buy_qv', math.nan)) or
           not 0 <= r['taker_buy_qv'] <= r['qv'] for r in recent):
        return None, 'missing_or_invalid_30_day_active_buy_fields'
    quote = st.median(r['qv'] for r in recent)
    if quote < min_median_quote:
        return None, 'median_daily_quote_below_5m' if min_median_quote == 5e6 else 'median_daily_quote_below_research_threshold'
    closes = [r['c'] for r in rows]
    p = closes[-1]
    sma30, sma90 = st.mean(closes[-30:]), st.mean(closes[-90:])
    r30, r90 = p / closes[-31] - 1, p / closes[0] - 1
    btc30 = benchmark[-1]['c'] / benchmark[-31]['c'] - 1
    peak, mdd = closes[0], 0.
    for price in closes:
        peak = max(peak, price)
        mdd = max(mdd, 1 - price / peak)
    vol = st.stdev(math.log(b / a) for a, b in zip(closes, closes[1:])) * math.sqrt(365)
    buy, total = sum(r['taker_buy_qv'] for r in recent), sum(r['qv'] for r in recent)
    share = buy / total
    parts = {
        'trend': 10 * (.4 * (p > sma30) + .3 * (sma30 > sma90) + .3 * clip(.5 + r90 / .6)),
        'relative_strength': 10 * clip(.5 + (r30 - btc30) / .4),
        'spot_buying': 10 * clip(.5 + (share - .5) / .1),
        'risk': 10 * (.5 * clip(1 - mdd / .5) + .5 * clip(1 - vol / 1.5)),
        'liquidity': 10 * clip(math.log10(quote / 5e6) / 2),
    }
    extension = p / sma30 - 1
    penalty = min(1., max(0., extension - .15) / .25)
    raw = max(0., sum(parts[k] * w for k, w in WEIGHTS.items()) - penalty)
    reasons = []
    if raw < 6: reasons.append('score_below_6')
    if p <= sma30: reasons.append('close_not_above_sma30')
    if r30 <= 0: reasons.append('30d_return_not_positive')
    if mdd > .4: reasons.append('90d_close_drawdown_above_40pct')
    return dict(symbol=symbol, as_of=end + DAY, score=raw, display_score=f'{raw:.1f}', components=parts,
                penalty=penalty, close=p, sma30=sma30, sma90=sma90, return_30d=r30, return_90d=r90,
                excess_btc_30d=r30-btc30, max_drawdown_90d=mdd, annualized_vol_90d=vol,
                median_quote_30d=quote, taker_buy_quote_30d=buy, total_quote_30d=total,
                active_buy_share_30d=share, extension_sma30=extension, qualified=not reasons,
                exclusion_reasons=reasons, stance='Wait for buying/price confirmation' if share < .5 or penalty else 'Research candidate, not an entry order'), None


def rank(rows):
    return sorted(rows, key=lambda r: (-r['score'], r['symbol']))


def sustained_score(symbol, series, btc, end):
    result, error = score(symbol, series, btc, end, min_median_quote=1e6)
    if error:
        return result, error
    recent = window(series, end, 30)
    result['median_quote_7d'] = st.median(r['qv'] for r in recent[-7:])
    result['days_above_500k_30d'] = sum(r['qv'] >= 5e5 for r in recent)
    if result['median_quote_7d'] < 1e6 or result['days_above_500k_30d'] < 24:
        return None, 'insufficient_sustained_turnover'
    return result, None


def relative_ratings(rows):
    """Strict empirical percentile over ALL scoreable peers, before risk gates.

    Ties receive equal ratings. No rank stretching over just the published ten.
    """
    n = len(rows)
    for row in rows:
        lower = sum(other['score'] < row['score'] for other in rows)
        row.update(relative_score=10*lower/n if n > 1 else None,
                   relative_lower_count=lower, relative_pool_n=n)


def outcome(series, decision, days):
    rows = [series.get(decision + i * DAY) for i in range(days)]
    if not all(r is not None and valid_bar(r) for r in rows):
        return None
    entry = rows[0]['o']
    return dict(entry_open=entry, exit_close=rows[-1]['c'], gross_return=rows[-1]['c']/entry-1,
                net_reference_return=rows[-1]['c']/entry-1-.002,
                mae=min(r['l']/entry-1 for r in rows), mfe=max(r['h']/entry-1 for r in rows))


def sensitivity(rows):
    """Leave one factor out, same originally qualified pool, not statistical CI."""
    ranges = {r['symbol']: [i+1] for i, r in enumerate(rank(rows))}
    for omitted, weight in WEIGHTS.items():
        ordered = sorted(rows, key=lambda r: (
            -max(0, sum(r['components'][k]*w for k,w in WEIGHTS.items() if k != omitted)/(1-weight)-r['penalty']), r['symbol']))
        for i, row in enumerate(ordered): ranges[row['symbol']].append(i+1)
    return {s: [min(v), max(v)] for s,v in ranges.items()}


def mean_interval(values, seed=17092026):
    """Descriptive 3-observation circular block bootstrap; overlapping months remain."""
    if len(values) < 12: return None
    rng = random.Random(seed)
    draws=[]
    for _ in range(2000):
        sample=[]
        while len(sample)<len(values):
            start=rng.randrange(len(values))
            sample.extend(values[(start+i)%len(values)] for i in range(3))
        draws.append(st.mean(sample[:len(values)]))
    draws.sort()
    def quantile(p):
        i=(len(draws)-1)*p;lo=int(i)
        return draws[lo]+(draws[min(lo+1,len(draws)-1)]-draws[lo])*(i-lo)
    return [quantile(.025),quantile(.975)]