from __future__ import annotations

"""
Batch 12E — Rolling Upcoming Form Layer

Computes game-based (5g / 10g) rolling batter and pitcher form metrics and
translates them into bounded additive probability adjustments for the UPCOMING
game engine only.

Design principles:
- Returns absolute rolling values; deltas are computed in the adjustment function
  against stable batter_features / pitcher_row baselines (NOT recomputed from the
  same narrow window).
- Sample-aware: weak windows (< 2 games) produce zero adjustment.
- 10g window used as confirmation/dampening of 5g signal.
- Pitcher-side adjustments scaled by pitcher_rolling_confidence (match quality
  × sample availability).
- Hard-capped adjustments; no runaway boosts.
- Pull/direction metrics SKIPPED (spray_angle not in normalized statcast).
- Zone/heart-rate metrics SKIPPED (not in normalized statcast).
"""

import logging
import re
import unicodedata
from datetime import date, datetime
from typing import Any

import pandas as pd

logger = logging.getLogger(__name__)

# ---------------------------------------------------------------------------
# Shared helpers (same barrel definition and utils as batter_trend_model)
# ---------------------------------------------------------------------------


def _parse_reference_date(reference_date: Any) -> date | None:
    if reference_date is None:
        return None
    if isinstance(reference_date, datetime):
        return reference_date.date()
    if isinstance(reference_date, date):
        return reference_date
    if isinstance(reference_date, str):
        for fmt in ("%Y-%m-%d", "%Y-%m-%dT%H:%M:%SZ", "%Y-%m-%dT%H:%M:%S"):
            try:
                return datetime.strptime(reference_date[:19], fmt).date()
            except ValueError:
                continue
    return None


def _percentile(series: pd.Series, q: float) -> float | None:
    numeric = pd.to_numeric(series, errors="coerce").dropna()
    if len(numeric) < 5:
        return None
    return float(numeric.quantile(q))


def _safe_mean(series: pd.Series) -> float | None:
    numeric = pd.to_numeric(series, errors="coerce").dropna()
    if len(numeric) < 5:
        return None
    return float(numeric.mean())


def _barrel_rate(launch_speed: pd.Series, launch_angle: pd.Series) -> float | None:
    valid = pd.DataFrame(
        {
            "ls": pd.to_numeric(launch_speed, errors="coerce"),
            "la": pd.to_numeric(launch_angle, errors="coerce"),
        }
    ).dropna()
    if len(valid) < 5:
        return None
    mask = (
        ((valid["ls"] >= 98) & (valid["la"].between(26, 30)))
        | ((valid["ls"] >= 99) & (valid["la"].between(25, 31)))
        | ((valid["ls"] >= 100) & (valid["la"].between(23, 33)))
        | ((valid["ls"] >= 102) & (valid["la"].between(20, 35)))
    )
    return float(mask.mean())


def _safe_rate_from_la(
    launch_angle: pd.Series,
    lo: float,
    hi: float | None = None,
) -> float | None:
    """Fraction of non-null LA rows where lo <= la < hi (or la >= lo if hi is None)."""
    la = pd.to_numeric(launch_angle, errors="coerce").dropna()
    if len(la) < 5:
        return None
    if hi is None:
        return float((la >= lo).mean())
    return float(((la >= lo) & (la < hi)).mean())


def _n_games(df: pd.DataFrame) -> int:
    """Count unique game_pk values in a slice; fall back to row-count heuristic."""
    if "game_pk" in df.columns:
        return int(df["game_pk"].nunique())
    return len(df)


# ---------------------------------------------------------------------------
# Game-window helper
# ---------------------------------------------------------------------------


def _game_window_df(player_df: pd.DataFrame, ref: date, n_games: int) -> pd.DataFrame:
    """
    Return rows for the last `n_games` unique games before `ref` (exclusive).

    Sorting is by `game_date` descending; unique `game_pk` values are taken in
    that order.  Falls back to the last N×25 rows (rough PA estimate) if
    `game_pk` is unavailable.
    """
    if player_df.empty:
        return player_df.iloc[0:0]

    if "game_date" not in player_df.columns:
        return player_df.iloc[0:0]

    game_dates = pd.to_datetime(player_df["game_date"], errors="coerce")
    cutoff = pd.Timestamp(ref)
    before_ref = player_df[game_dates < cutoff].copy()

    if before_ref.empty:
        return before_ref

    before_ref["_gd"] = pd.to_datetime(before_ref["game_date"], errors="coerce")

    if "game_pk" in before_ref.columns:
        before_ref["_gpk"] = pd.to_numeric(before_ref["game_pk"], errors="coerce")
        sorted_games = (
            before_ref.groupby("_gpk")["_gd"]
            .max()
            .sort_values(ascending=False)
            .head(n_games)
            .index.tolist()
        )
        result = before_ref[before_ref["_gpk"].isin(sorted_games)].drop(
            columns=["_gd", "_gpk"], errors="ignore"
        )
        return result

    # Fallback: no game_pk — take last n_games*25 rows sorted by date
    fallback = before_ref.sort_values("_gd", ascending=False).head(n_games * 25)
    return fallback.drop(columns=["_gd"], errors="ignore")


# ---------------------------------------------------------------------------
# Pitcher name normalization (mirrors pitcher_adjustment.py)
# ---------------------------------------------------------------------------


def _normalize_name(name: str) -> str:
    text = str(name or "").strip().lower()
    text = unicodedata.normalize("NFKD", text)
    text = "".join(ch for ch in text if not unicodedata.combining(ch))
    text = text.replace(",", " ")
    text = re.sub(r"\s+", " ", text).strip()
    return text


def _name_variants(name: str) -> set[str]:
    normalized = _normalize_name(name)
    if not normalized:
        return set()
    parts = normalized.split()
    variants = {normalized}
    if len(parts) >= 2:
        first, last = parts[0], parts[-1]
        middle = " ".join(parts[1:-1]).strip()
        variants.add(f"{last} {first}".strip())
        if middle:
            variants.add(f"{last} {first} {middle}".strip())
    return variants


# ---------------------------------------------------------------------------
# Empty skeletons
# ---------------------------------------------------------------------------

_EMPTY_BATTER_ROLL: dict[str, Any] = {
    "batter_ev_5g": None,
    "batter_ev_10g": None,
    "batter_ev90_5g": None,
    "batter_ev90_10g": None,
    "batter_hard_hit_rate_5g": None,
    "batter_hard_hit_rate_10g": None,
    "batter_barrel_rate_5g": None,
    "batter_barrel_rate_10g": None,
    "batter_avg_launch_angle_5g": None,
    "batter_avg_launch_angle_10g": None,
    "batter_fb_rate_5g": None,
    "batter_fb_rate_10g": None,
    "batter_ld_rate_5g": None,
    "batter_gb_rate_5g": None,
    "batter_air_ball_rate_5g": None,
    "batter_hr_rate_5g": None,
    "batter_hr_rate_10g": None,
    # direction metrics deferred (spray_angle not in normalized statcast)
    "batter_pull_air_rate_5g": None,
    "batter_pulled_hard_air_rate_5g": None,
    "batter_pulled_barrel_rate_5g": None,
    "batter_games_in_window_5g": 0,
    "batter_games_in_window_10g": 0,
    "batter_recent_form_available": 0,
}

_EMPTY_PITCHER_ROLL: dict[str, Any] = {
    "pitcher_avg_release_speed_5g": None,
    "pitcher_avg_release_speed_10g": None,
    "pitcher_avg_release_spin_rate_5g": None,
    "pitcher_ev_allowed_5g": None,
    "pitcher_ev_allowed_10g": None,
    "pitcher_hard_hit_rate_allowed_5g": None,
    "pitcher_hard_hit_rate_allowed_10g": None,
    "pitcher_barrel_rate_allowed_5g": None,
    "pitcher_barrel_rate_allowed_10g": None,
    "pitcher_avg_launch_angle_allowed_5g": None,
    "pitcher_fb_rate_allowed_5g": None,
    "pitcher_ld_rate_allowed_5g": None,
    "pitcher_gb_rate_allowed_5g": None,
    "pitcher_hr_allowed_rate_5g": None,
    "pitcher_hr_allowed_rate_10g": None,
    "pitcher_games_in_window_5g": 0,
    "pitcher_games_in_window_10g": 0,
    "pitcher_recent_form_available": 0,
    "pitcher_rolling_confidence": 0.0,
}


# ---------------------------------------------------------------------------
# Public API — batter rolling form
# ---------------------------------------------------------------------------


def build_batter_rolling_form_row(
    statcast_df: pd.DataFrame,
    player_name: str,
    reference_date: Any = None,
) -> dict[str, Any]:
    """
    Compute game-based 5g / 10g rolling form metrics for *player_name*.

    Returns absolute rolling values only; delta vs. baseline is handled in
    compute_upcoming_rolling_adjustment() against stable batter_features values.
    """
    if statcast_df is None or statcast_df.empty:
        return dict(_EMPTY_BATTER_ROLL)

    ref = _parse_reference_date(reference_date)
    if ref is None:
        return dict(_EMPTY_BATTER_ROLL)

    try:
        player_df = statcast_df[
            statcast_df["player_name"].astype(str) == str(player_name)
        ].copy()
    except Exception:
        return dict(_EMPTY_BATTER_ROLL)

    if player_df.empty:
        return dict(_EMPTY_BATTER_ROLL)

    df5 = _game_window_df(player_df, ref, 5)
    df10 = _game_window_df(player_df, ref, 10)

    n5 = _n_games(df5)
    n10 = _n_games(df10)

    def _hr_rate(df: pd.DataFrame) -> float | None:
        if "events" not in df.columns or len(df) < 5:
            return None
        events = df["events"].dropna().astype(str)
        if events.empty:
            return None
        return float((events == "home_run").mean())

    def _hh_rate(df: pd.DataFrame) -> float | None:
        ls = pd.to_numeric(df.get("launch_speed", pd.Series(dtype=float)), errors="coerce").dropna()
        if len(ls) < 5:
            return None
        return float((ls >= 95).mean())

    ls5 = df5.get("launch_speed", pd.Series(dtype=float)) if not df5.empty else pd.Series(dtype=float)
    la5 = df5.get("launch_angle", pd.Series(dtype=float)) if not df5.empty else pd.Series(dtype=float)
    ls10 = df10.get("launch_speed", pd.Series(dtype=float)) if not df10.empty else pd.Series(dtype=float)
    la10 = df10.get("launch_angle", pd.Series(dtype=float)) if not df10.empty else pd.Series(dtype=float)

    return {
        "batter_ev_5g": _safe_mean(ls5),
        "batter_ev_10g": _safe_mean(ls10),
        "batter_ev90_5g": _percentile(ls5, 0.90),
        "batter_ev90_10g": _percentile(ls10, 0.90),
        "batter_hard_hit_rate_5g": _hh_rate(df5),
        "batter_hard_hit_rate_10g": _hh_rate(df10),
        "batter_barrel_rate_5g": _barrel_rate(ls5, la5),
        "batter_barrel_rate_10g": _barrel_rate(ls10, la10),
        "batter_avg_launch_angle_5g": _safe_mean(la5),
        "batter_avg_launch_angle_10g": _safe_mean(la10),
        "batter_fb_rate_5g": _safe_rate_from_la(la5, 25.0),
        "batter_fb_rate_10g": _safe_rate_from_la(la10, 25.0),
        "batter_ld_rate_5g": _safe_rate_from_la(la5, 10.0, 25.0),
        "batter_gb_rate_5g": _safe_rate_from_la(la5, -90.0, 10.0),
        "batter_air_ball_rate_5g": _safe_rate_from_la(la5, 10.0),
        "batter_hr_rate_5g": _hr_rate(df5),
        "batter_hr_rate_10g": _hr_rate(df10),
        # direction metrics deferred
        "batter_pull_air_rate_5g": None,
        "batter_pulled_hard_air_rate_5g": None,
        "batter_pulled_barrel_rate_5g": None,
        "batter_games_in_window_5g": n5,
        "batter_games_in_window_10g": n10,
        "batter_recent_form_available": 1 if n5 >= 4 else 0,
    }


# ---------------------------------------------------------------------------
# Public API — pitcher rolling form
# ---------------------------------------------------------------------------


def build_pitcher_rolling_form_row(
    statcast_df: pd.DataFrame,
    pitcher_name: str | None = None,
    pitcher_id: int | None = None,
    reference_date: Any = None,
) -> dict[str, Any]:
    """
    Compute game-based 5g / 10g rolling form metrics for a pitcher.

    Follows the same fuzzy-name-match pattern as pitcher_adjustment.py.
    pitcher_rolling_confidence reflects match quality × sample availability.
    """
    if statcast_df is None or statcast_df.empty:
        return dict(_EMPTY_PITCHER_ROLL)

    ref = _parse_reference_date(reference_date)
    if ref is None:
        return dict(_EMPTY_PITCHER_ROLL)

    pitcher_name = str(pitcher_name or "").strip()

    df = pd.DataFrame()
    match_quality = "none"

    # Attempt 1: pitcher ID column (present in some CSVs)
    if pitcher_id is not None and "pitcher" in statcast_df.columns:
        try:
            numeric_ids = pd.to_numeric(statcast_df["pitcher"], errors="coerce")
            df = statcast_df[numeric_ids == int(pitcher_id)].copy()
            if not df.empty:
                match_quality = "id"
        except Exception:
            df = pd.DataFrame()

    # Attempt 2: exact / variant name match on player_name
    if df.empty and pitcher_name and "player_name" in statcast_df.columns:
        variants = _name_variants(pitcher_name)
        normalized_series = statcast_df["player_name"].astype(str).map(_normalize_name)
        mask = normalized_series.isin(variants)
        df = statcast_df[mask].copy()
        if not df.empty:
            match_quality = "exact"

    # Attempt 3: loose contains-style match
    if df.empty and pitcher_name and "player_name" in statcast_df.columns:
        parts = _normalize_name(pitcher_name).split()
        if len(parts) >= 2:
            first, last = parts[0], parts[-1]
            normalized_series = statcast_df["player_name"].astype(str).map(_normalize_name)
            loose_mask = normalized_series.apply(
                lambda n: isinstance(n, str) and first in n and last in n
            )
            df = statcast_df[loose_mask].copy()
            if not df.empty:
                match_quality = "loose"

    if df.empty:
        return dict(_EMPTY_PITCHER_ROLL)

    df5 = _game_window_df(df, ref, 5)
    df10 = _game_window_df(df, ref, 10)

    n5 = _n_games(df5)
    n10 = _n_games(df10)

    # pitcher_rolling_confidence: match quality × sample scale
    sample_scale_5g = (
        0.0 if n5 < 2
        else 0.4 if n5 <= 3
        else 0.7 if n5 == 4
        else 1.0
    )
    match_scale = {
        "id": 1.0,
        "exact": 1.0,
        "loose": 0.4,
        "none": 0.0,
    }.get(match_quality, 0.0)
    confidence = round(match_scale * sample_scale_5g, 3)

    def _hh_rate(df: pd.DataFrame) -> float | None:
        ls = pd.to_numeric(df.get("launch_speed", pd.Series(dtype=float)), errors="coerce").dropna()
        if len(ls) < 5:
            return None
        return float((ls >= 95).mean())

    def _hr_rate_allowed(df: pd.DataFrame) -> float | None:
        if "events" not in df.columns or len(df) < 5:
            return None
        events = df["events"].dropna().astype(str)
        if events.empty:
            return None
        return float((events == "home_run").mean())

    ls5 = df5.get("launch_speed", pd.Series(dtype=float)) if not df5.empty else pd.Series(dtype=float)
    la5 = df5.get("launch_angle", pd.Series(dtype=float)) if not df5.empty else pd.Series(dtype=float)
    ls10 = df10.get("launch_speed", pd.Series(dtype=float)) if not df10.empty else pd.Series(dtype=float)
    la10 = df10.get("launch_angle", pd.Series(dtype=float)) if not df10.empty else pd.Series(dtype=float)
    rs5 = df5.get("release_speed", pd.Series(dtype=float)) if not df5.empty else pd.Series(dtype=float)
    rs10 = df10.get("release_speed", pd.Series(dtype=float)) if not df10.empty else pd.Series(dtype=float)
    spin5 = df5.get("release_spin_rate", pd.Series(dtype=float)) if not df5.empty else pd.Series(dtype=float)

    return {
        "pitcher_avg_release_speed_5g": _safe_mean(rs5),
        "pitcher_avg_release_speed_10g": _safe_mean(rs10),
        "pitcher_avg_release_spin_rate_5g": _safe_mean(spin5),
        "pitcher_ev_allowed_5g": _safe_mean(ls5),
        "pitcher_ev_allowed_10g": _safe_mean(ls10),
        "pitcher_hard_hit_rate_allowed_5g": _hh_rate(df5),
        "pitcher_hard_hit_rate_allowed_10g": _hh_rate(df10),
        "pitcher_barrel_rate_allowed_5g": _barrel_rate(ls5, la5),
        "pitcher_barrel_rate_allowed_10g": _barrel_rate(ls10, la10),
        "pitcher_avg_launch_angle_allowed_5g": _safe_mean(la5),
        "pitcher_fb_rate_allowed_5g": _safe_rate_from_la(la5, 25.0),
        "pitcher_ld_rate_allowed_5g": _safe_rate_from_la(la5, 10.0, 25.0),
        "pitcher_gb_rate_allowed_5g": _safe_rate_from_la(la5, -90.0, 10.0),
        "pitcher_hr_allowed_rate_5g": _hr_rate_allowed(df5),
        "pitcher_hr_allowed_rate_10g": _hr_rate_allowed(df10),
        "pitcher_games_in_window_5g": n5,
        "pitcher_games_in_window_10g": n10,
        "pitcher_recent_form_available": 1 if n5 >= 4 else 0,
        "pitcher_rolling_confidence": confidence,
    }


# ---------------------------------------------------------------------------
# Helpers for adjustment function
# ---------------------------------------------------------------------------


def _safe_delta(rolling_val: Any, baseline_val: Any) -> float | None:
    """rolling - baseline; returns None if either is None."""
    if rolling_val is None or baseline_val is None:
        return None
    try:
        return float(rolling_val) - float(baseline_val)
    except (TypeError, ValueError):
        return None


def _clamp(value: float, lo: float, hi: float) -> float:
    return max(lo, min(hi, value))


def _10g_confirmation_scale(delta_5g: float | None, delta_10g: float | None, threshold: float) -> float:
    """
    1.0 if 10g confirms 5g direction or is None (neutral).
    0.5 if 10g conflicts with 5g direction.
    """
    if delta_5g is None or delta_10g is None:
        return 1.0
    aligned = (delta_5g > threshold) == (delta_10g > threshold)
    return 1.0 if aligned else 0.5


def _sample_scale(n_games: int) -> float:
    if n_games < 4:   return 0.0   # raised gate to match recent_form_available (n >= 4)
    if n_games == 4:  return 0.7
    return 1.0


# ---------------------------------------------------------------------------
# Public API — rolling adjustment
# ---------------------------------------------------------------------------


def compute_upcoming_rolling_adjustment(
    batter_roll: dict[str, Any],
    pitcher_roll: dict[str, Any],
    batter_features: dict[str, Any],
    pitcher_row: dict[str, Any],
) -> dict[str, Any]:
    """
    Compute bounded additive probability adjustments from rolling form.

    Deltas are computed against the stable batter_features / pitcher_row
    baselines (not recomputed from the narrow rolling window).

    Returns a dict with rolling_hit_adjustment, rolling_hr_adjustment,
    rolling_tb2p_adjustment, scores, tags (pipe-delimited string), and
    pitcher_rolling_confidence.
    """
    batter_n5 = int(batter_roll.get("batter_games_in_window_5g") or 0)
    pitcher_n5 = int(pitcher_roll.get("pitcher_games_in_window_5g") or 0)
    pitcher_confidence = float(pitcher_roll.get("pitcher_rolling_confidence") or 0.0)

    batter_scale = _sample_scale(batter_n5)
    pitcher_n5_scale = _sample_scale(pitcher_n5)
    pitcher_scale = pitcher_confidence * pitcher_n5_scale

    # ------------------------------------------------------------------
    # Compute deltas vs stable engine baselines
    # ------------------------------------------------------------------

    # Batter deltas
    ev90_delta_5g = _safe_delta(batter_roll.get("batter_ev90_5g"), batter_features.get("ev90"))
    ev90_delta_10g = _safe_delta(batter_roll.get("batter_ev90_10g"), batter_features.get("ev90"))
    barrel_delta_5g = _safe_delta(batter_roll.get("batter_barrel_rate_5g"), batter_features.get("barrel_rate"))
    barrel_delta_10g = _safe_delta(batter_roll.get("batter_barrel_rate_10g"), batter_features.get("barrel_rate"))
    hh_delta_5g = _safe_delta(batter_roll.get("batter_hard_hit_rate_5g"), batter_features.get("hard_hit_rate"))
    la_delta_5g = _safe_delta(batter_roll.get("batter_avg_launch_angle_5g"), batter_features.get("avg_launch_angle"))
    air_ball_5g = batter_roll.get("batter_air_ball_rate_5g")
    air_ball_baseline = batter_features.get("air_ball_rate")
    air_ball_delta_5g = _safe_delta(air_ball_5g, air_ball_baseline)

    # Pitcher deltas vs stable pitcher_row baselines
    velo_delta_5g = _safe_delta(pitcher_roll.get("pitcher_avg_release_speed_5g"), pitcher_row.get("avg_release_speed"))
    ev_allowed_delta_5g = _safe_delta(pitcher_roll.get("pitcher_ev_allowed_5g"), pitcher_row.get("ev_allowed"))
    ev_allowed_delta_10g = _safe_delta(pitcher_roll.get("pitcher_ev_allowed_10g"), pitcher_row.get("ev_allowed"))
    barrel_allowed_delta_5g = _safe_delta(pitcher_roll.get("pitcher_barrel_rate_allowed_5g"), pitcher_row.get("barrel_rate_allowed"))
    barrel_allowed_delta_10g = _safe_delta(pitcher_roll.get("pitcher_barrel_rate_allowed_10g"), pitcher_row.get("barrel_rate_allowed"))
    hh_allowed_delta_5g = _safe_delta(pitcher_roll.get("pitcher_hard_hit_rate_allowed_5g"), pitcher_row.get("hard_hit_rate_allowed"))

    # ------------------------------------------------------------------
    # Batter form score
    # ------------------------------------------------------------------
    batter_score = 0.0
    active_batter_tags: list[str] = []

    if ev90_delta_5g is not None:
        conf_10g = _10g_confirmation_scale(ev90_delta_5g, ev90_delta_10g, 2.0)
        if ev90_delta_5g > 2.0:
            batter_score += 0.25 * conf_10g
            active_batter_tags.append("batter_ev90_surge")
        elif ev90_delta_5g < -2.0:
            batter_score -= 0.25 * conf_10g
            active_batter_tags.append("batter_ev90_decline")

    if barrel_delta_5g is not None:
        conf_10g = _10g_confirmation_scale(barrel_delta_5g, barrel_delta_10g, 0.03)
        if barrel_delta_5g > 0.03:
            batter_score += 0.40 * conf_10g
            active_batter_tags.append("batter_barrel_spike")
        elif barrel_delta_5g < -0.03:
            batter_score -= 0.40 * conf_10g
            active_batter_tags.append("batter_barrel_drop")

    if hh_delta_5g is not None and hh_delta_5g > 0.05:
        batter_score += 0.20
        active_batter_tags.append("batter_hard_hit_rising")

    if (
        la_delta_5g is not None
        and batter_roll.get("batter_avg_launch_angle_5g") is not None
        and 20.0 < float(batter_roll["batter_avg_launch_angle_5g"]) < 30.0
        and la_delta_5g > 3.0
    ):
        batter_score += 0.20
        active_batter_tags.append("batter_la_optimizing")

    if (
        air_ball_5g is not None
        and air_ball_delta_5g is not None
        and float(air_ball_5g) > 0.45
        and air_ball_delta_5g > 0.05
    ):
        batter_score += 0.15
        active_batter_tags.append("batter_air_ball_spike")

    batter_score = _clamp(batter_score * batter_scale, -1.0, 1.0)

    # ------------------------------------------------------------------
    # Pitcher form score
    # ------------------------------------------------------------------
    pitcher_score = 0.0
    active_pitcher_tags: list[str] = []

    if velo_delta_5g is not None:
        if velo_delta_5g < -3.0:
            pitcher_score += 0.50  # -1.5 and -3.0 contributions combined
            active_pitcher_tags.append("pitcher_velo_decline_hard")
        elif velo_delta_5g < -1.5:
            pitcher_score += 0.30
            active_pitcher_tags.append("pitcher_velo_decline")

    if ev_allowed_delta_5g is not None:
        conf_10g = _10g_confirmation_scale(ev_allowed_delta_5g, ev_allowed_delta_10g, 2.0)
        if ev_allowed_delta_5g > 2.0:
            pitcher_score += 0.30 * conf_10g
            active_pitcher_tags.append("pitcher_ev_allowed_spiking")

    if barrel_allowed_delta_5g is not None:
        conf_10g = _10g_confirmation_scale(barrel_allowed_delta_5g, barrel_allowed_delta_10g, 0.03)
        if barrel_allowed_delta_5g > 0.03:
            pitcher_score += 0.40 * conf_10g
            active_pitcher_tags.append("pitcher_barrel_allowed_spiking")

    if hh_allowed_delta_5g is not None and hh_allowed_delta_5g > 0.05:
        pitcher_score += 0.20
        active_pitcher_tags.append("pitcher_hard_hit_allowed_rising")

    # Pitcher sharp: velo up + EV allowed down + barrel allowed down
    pitcher_sharp = (
        velo_delta_5g is not None and velo_delta_5g > 1.5
        and ev_allowed_delta_5g is not None and ev_allowed_delta_5g < -2.0
        and barrel_allowed_delta_5g is not None and barrel_allowed_delta_5g < -0.03
    )
    if pitcher_sharp:
        pitcher_score -= 0.35
        active_pitcher_tags.append("pitcher_sharp_recently")

    pitcher_score = _clamp(pitcher_score * pitcher_scale, -1.0, 1.0)

    # ------------------------------------------------------------------
    # Combined score and adjustments
    # ------------------------------------------------------------------
    combined = _clamp(batter_score + pitcher_score, -1.0, 1.0)

    rolling_hr_adjustment   = _clamp(combined * 0.012, -0.012, 0.012)
    rolling_hit_adjustment  = _clamp(combined * 0.010, -0.010, 0.010)
    rolling_tb2p_adjustment = _clamp(combined * 0.011, -0.011, 0.011)

    adjustment_applied = abs(combined) > 0.05

    # Compact pipe-delimited reason tags (up to 3 most active)
    all_tags = (active_batter_tags + active_pitcher_tags)[:3]
    reason_tags_str = "|".join(all_tags)

    return {
        "rolling_hit_adjustment": round(rolling_hit_adjustment, 5),
        "rolling_hr_adjustment": round(rolling_hr_adjustment, 5),
        "rolling_tb2p_adjustment": round(rolling_tb2p_adjustment, 5),
        "rolling_batter_form_score": round(batter_score, 4),
        "rolling_pitcher_form_score": round(pitcher_score, 4),
        "rolling_combined_form_score": round(combined, 4),
        "rolling_adjustment_applied": adjustment_applied,
        "rolling_adjustment_reason_tags": reason_tags_str,
        "pitcher_rolling_confidence": pitcher_confidence,
    }