engine_rust_src/src/core/mcts.rs

#[cfg(feature = "extension-module")]
use pyo3::prelude::*;
use rand::prelude::*;
use rand::rngs::SmallRng;
use rand::SeedableRng;
use crate::core::heuristics::Heuristic;
use crate::core::logic::{
    GameState, CardDatabase, LiveCard, Phase,
    FLAG_DRAW, FLAG_SEARCH, FLAG_RECOVER, FLAG_BUFF, FLAG_CHARGE,
    FLAG_TEMPO, FLAG_REDUCE, FLAG_BOOST, FLAG_TRANSFORM, FLAG_WIN_COND
};
use std::f32;
#[cfg(feature = "parallel")]
use rayon::prelude::*;
use std::collections::HashMap;
#[cfg(feature = "nn")]
use ort::session::Session;
#[cfg(feature = "nn")]
use std::sync::{Arc, Mutex};
use std::time::Duration;

#[derive(Default, Clone, Copy)]
pub struct MCTSProfiler {
    pub determinization: Duration,
    pub selection: Duration,
    pub expansion: Duration,
    pub simulation: Duration,
    pub backpropagation: Duration,
}

impl MCTSProfiler {
    pub fn merge(&mut self, other: &Self) {
        self.determinization += other.determinization;
        self.selection += other.selection;
        self.expansion += other.expansion;
        self.simulation += other.simulation;
        self.backpropagation += other.backpropagation;
    }

    pub fn print(&self, total: Duration) {
        let total_secs = total.as_secs_f64();
        if total_secs == 0.0 { return; }
        println!("[MCTS Profile] Breakdown:");
        let items = [
            ("Determinization", self.determinization),
            ("Selection", self.selection),
            ("Expansion", self.expansion),
            ("Simulation", self.simulation),
            ("Backpropagation", self.backpropagation),
        ];

        for (name, dur) in items {
            let secs = dur.as_secs_f64();
            println!("  - {:<16}: {:>8.3}s ({:>6.1}%)", name, secs, (secs / total_secs) * 100.0);
        }
    }
}

struct Node {
    visit_count: u32,
    value_sum: f32,
    player_just_moved: u8,
    untried_actions: Vec<i32>,
    children: Vec<(i32, usize)>, // (Action, NodeIndex in Arena)
    parent: Option<usize>,
    parent_action: i32,
}

pub struct MCTS {
    nodes: Vec<Node>,
    rng: SmallRng,
    unseen_buffer: Vec<u16>,
    legal_buffer: Vec<i32>,
    reusable_state: GameState,
}

#[cfg_attr(feature = "extension-module", pyclass(eq, eq_int))]
#[derive(Debug, Clone, Copy, PartialEq)]
pub enum SearchHorizon {
    GameEnd,
    TurnEnd,
}

#[cfg_attr(feature = "extension-module", pyclass(eq, eq_int))]
#[derive(Debug, Clone, Copy, PartialEq)]
pub enum EvalMode {
    Normal,
    Solitaire,
    Blind,
}

impl MCTS {
    pub fn new() -> Self {
        Self {
            nodes: Vec::with_capacity(1000),
            rng: SmallRng::from_os_rng(),
            unseen_buffer: Vec::with_capacity(60),
            legal_buffer: Vec::with_capacity(32),
            reusable_state: GameState::default(),
        }
    }

    pub fn search_parallel(&self, root_state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32, horizon: SearchHorizon, heuristic: &dyn Heuristic, shuffle_self: bool) -> Vec<(i32, f32, u32)> {
        let start_overall = std::time::Instant::now();
        
        #[cfg(feature = "parallel")]
        let num_threads = rayon::current_num_threads().max(1);
        #[cfg(not(feature = "parallel"))]
        let num_threads = 1;

        let sims_per_thread = if num_sims > 0 { (num_sims + num_threads - 1) / num_threads } else { 0 };

        // Collect results
        #[cfg(feature = "parallel")]
        let results: Vec<(Vec<(i32, f32, u32)>, MCTSProfiler)> = (0..num_threads).into_par_iter().map(|_| {
            let mut mcts = MCTS::new();
            mcts.search_custom(root_state, db, sims_per_thread, timeout_sec, horizon, heuristic, shuffle_self, true)
        }).collect();

        #[cfg(not(feature = "parallel"))]
        let results: Vec<(Vec<(i32, f32, u32)>, MCTSProfiler)> = vec![{
             let mut mcts = MCTS::new();
             mcts.search_custom(root_state, db, num_sims, timeout_sec, horizon, heuristic, shuffle_self, true)
        }];

        // Merge results
        let mut agg_map: HashMap<i32, (f32, u32)> = HashMap::new();
        let mut total_visits = 0;
        let mut agg_profile = MCTSProfiler::default();

        for (res, profile) in results {
            agg_profile.merge(&profile);
            for (action, score, visits) in res {
                let entry = agg_map.entry(action).or_insert((0.0, 0));
                let total_value = score * visits as f32;
                entry.0 += total_value;
                entry.1 += visits;
                total_visits += visits;
            }
        }

        // Logging Speed
        let duration = start_overall.elapsed();
        let sims_per_sec = total_visits as f64 / duration.as_secs_f64();
        if total_visits > 100 {
             println!("[MCTS] Completed {} sims in {:.3}s ({:.0} sims/s)", total_visits, duration.as_secs_f64(), sims_per_sec);
             agg_profile.print(duration);
        }

        let mut stats: Vec<(i32, f32, u32)> = agg_map.into_iter().map(|(action, (sum_val, visits))| {
            if visits > 0 {
                (action, sum_val / visits as f32, visits)
            } else {
                (action, 0.0, 0)
            }
        }).collect();

        stats.sort_by_key(|&(_, _, v)| std::cmp::Reverse(v));
        stats
    }

    pub fn search_parallel_mode(&self, root_state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32, horizon: SearchHorizon, eval_mode: EvalMode) -> Vec<(i32, f32, u32)> {
        let start_overall = std::time::Instant::now();
        
        #[cfg(feature = "parallel")]
        let num_threads = rayon::current_num_threads().max(1);
        #[cfg(not(feature = "parallel"))]
        let num_threads = 1;

        let sims_per_thread = if num_sims > 0 { (num_sims + num_threads - 1) / num_threads } else { 0 };

        // Collect results
        #[cfg(feature = "parallel")]
        let results: Vec<(Vec<(i32, f32, u32)>, MCTSProfiler)> = (0..num_threads).into_par_iter().map(|_| {
            let mut mcts = MCTS::new();
            mcts.search_mode(root_state, db, sims_per_thread, timeout_sec, horizon, eval_mode)
        }).collect();

        #[cfg(not(feature = "parallel"))]
        let results: Vec<(Vec<(i32, f32, u32)>, MCTSProfiler)> = vec![{
             let mut mcts = MCTS::new();
             mcts.search_mode(root_state, db, num_sims, timeout_sec, horizon, eval_mode)
        }];

        // Merge results
        let mut agg_map: HashMap<i32, (f32, u32)> = HashMap::new();
        let mut total_visits = 0;
        let mut agg_profile = MCTSProfiler::default();

        for (res, profile) in results {
            agg_profile.merge(&profile);
            for (action, score, visits) in res {
                let entry = agg_map.entry(action).or_insert((0.0, 0));
                let total_value = score * visits as f32;
                entry.0 += total_value;
                entry.1 += visits;
                total_visits += visits;
            }
        }

        // Logging Speed
        let duration = start_overall.elapsed();
        let sims_per_sec = total_visits as f64 / duration.as_secs_f64();
        if total_visits > 100 {
             println!("[MCTS Mode] Completed {} sims in {:.3}s ({:.0} sims/s)", total_visits, duration.as_secs_f64(), sims_per_sec);
             agg_profile.print(duration);
        }

        let mut stats: Vec<(i32, f32, u32)> = agg_map.into_iter().map(|(action, (sum_val, visits))| {
            if visits > 0 {
                (action, sum_val / visits as f32, visits)
            } else {
                (action, 0.0, 0)
            }
        }).collect();

        stats.sort_by_key(|&(_, _, v)| std::cmp::Reverse(v));
        stats
    }

    pub fn search(&mut self, root_state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32, horizon: SearchHorizon, heuristic: &dyn Heuristic) -> (Vec<(i32, f32, u32)>, MCTSProfiler) {
        self.search_custom(root_state, db, num_sims, timeout_sec, horizon, heuristic, false, true)
    }

    pub fn search_custom(&mut self, root_state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32, horizon: SearchHorizon, heuristic: &dyn Heuristic, shuffle_self: bool, enable_rollout: bool) -> (Vec<(i32, f32, u32)>, MCTSProfiler) {
        self.run_mcts_config(root_state, db, num_sims, timeout_sec, horizon, shuffle_self, enable_rollout, |state, _db| {
            if state.is_terminal() {
                 match state.get_winner() {
                    0 => 1.0,
                    1 => 0.0,
                    _ => 0.5,
                }
            } else {
                heuristic.evaluate(state, db, root_state.players[0].score, root_state.players[1].score)
            }
        })
    }

    pub fn search_mode(&mut self, root_state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32, horizon: SearchHorizon, eval_mode: EvalMode) -> (Vec<(i32, f32, u32)>, MCTSProfiler) {
        // Pre-calculate deck expectations for optimization
        // P0 (Me): Use current deck
        let p0_stats = Self::calculate_deck_expectations(&root_state.players[0].deck, db);

        // P1 (Opponent): Use Hand + Deck (Unseen)
        let mut p1_unseen = root_state.players[1].hand.clone();
        p1_unseen.extend(root_state.players[1].deck.iter().cloned());
        let p1_stats = Self::calculate_deck_expectations(&p1_unseen, db);

        self.run_mcts_config(root_state, db, num_sims, timeout_sec, horizon, eval_mode == EvalMode::Blind, true, |state, _db| {
            if state.is_terminal() {
                if eval_mode == EvalMode::Solitaire {
                     match state.get_winner() {
                        0 => 1.0,
                        1 => 0.0,
                        _ => 0.5,
                    }
                } else {
                    match state.get_winner() {
                        0 => 1.0,
                        1 => 0.0,
                        _ => 0.5,
                    }
                }
            } else {
                Self::heuristic_eval(state, db, root_state.players[0].score, root_state.players[1].score, eval_mode, Some(p0_stats), Some(p1_stats))
            }
        })
    }


    fn run_mcts_config<F>(&mut self, root_state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32, horizon: SearchHorizon, shuffle_self: bool, enable_rollout: bool, mut eval_fn: F) -> (Vec<(i32, f32, u32)>, MCTSProfiler)
    where F: FnMut(&GameState, &CardDatabase) -> f32
    {
        self.nodes.clear();
        let start_time = std::time::Instant::now();
        let timeout = if timeout_sec > 0.0 { Some(std::time::Duration::from_secs_f32(timeout_sec)) } else { None };
        let start_turn = root_state.turn;
        let mut profiler = MCTSProfiler::default();

        // Root Node
        let mut legal_indices: Vec<i32> = Vec::with_capacity(32);
        root_state.generate_legal_actions(db, &mut legal_indices);

        if legal_indices.is_empty() { return (vec![(0, 0.5, 0)], profiler); }
        if legal_indices.len() == 1 { return (vec![(legal_indices[0], 0.5, 1)], profiler); }

        self.nodes.push(Node {
            visit_count: 0,
            value_sum: 0.0,
            player_just_moved: 1 - root_state.current_player,
            untried_actions: legal_indices,
            children: Vec::new(),
            parent: None,
            parent_action: 0,
        });

        let mut sims_done = 0;
        loop {
            if num_sims > 0 && sims_done >= num_sims { break; }
            if let Some(to) = timeout {
                if start_time.elapsed() >= to { break; }
            }
            if num_sims == 0 && timeout.is_none() { break; } // Safety

            sims_done += 1;

            // 1. Setup & Determinization
            let t_setup = std::time::Instant::now();
            let mut node_idx = 0;
            self.reusable_state.copy_from(&root_state);
            let state = &mut self.reusable_state;
            state.silent = true;

            let me = root_state.current_player as usize;
            let opp = 1 - me;
            let opp_hand_len = state.players[opp].hand.len();

            self.unseen_buffer.clear();
            self.unseen_buffer.extend_from_slice(&state.players[opp].hand);
            self.unseen_buffer.extend_from_slice(&state.players[opp].deck);
            self.unseen_buffer.shuffle(&mut self.rng);

            state.players[opp].hand.copy_from_slice(&self.unseen_buffer[0..opp_hand_len]);
            state.players[opp].deck.copy_from_slice(&self.unseen_buffer[opp_hand_len..]);

            if shuffle_self {
                let mut my_deck = state.players[me].deck.clone();
                my_deck.shuffle(&mut self.rng);
                state.players[me].deck = my_deck;
            }
            profiler.determinization += t_setup.elapsed();

            // 2. Selection
            let t_selection = std::time::Instant::now();
            while self.nodes[node_idx].untried_actions.is_empty() && !self.nodes[node_idx].children.is_empty() {
                node_idx = Self::select_child(&self.nodes, node_idx);
                let action = self.nodes[node_idx].parent_action;
                let _ = state.step(db, action);
            }
            profiler.selection += t_selection.elapsed();

            // 3. Expansion
            let t_expansion = std::time::Instant::now();
            if !self.nodes[node_idx].untried_actions.is_empty() {
                let idx = self.rng.random_range(0..self.nodes[node_idx].untried_actions.len());
                let action = self.nodes[node_idx].untried_actions.swap_remove(idx);

                let actor = state.current_player;
                let _ = state.step(db, action);

                let mut new_legal_indices: Vec<i32> = Vec::with_capacity(32);
                state.generate_legal_actions(db, &mut new_legal_indices);

                let new_node = Node {
                    visit_count: 0,
                    value_sum: 0.0,
                    player_just_moved: actor,
                    untried_actions: new_legal_indices,
                    children: Vec::new(),
                    parent: Some(node_idx),
                    parent_action: action,
                };

                let new_idx = self.nodes.len();
                self.nodes.push(new_node);
                self.nodes[node_idx].children.push((action, new_idx));
                node_idx = new_idx;
            }
            profiler.expansion += t_expansion.elapsed();

            // 4. Simulation
            let t_simulation = std::time::Instant::now();
            let mut depth = 0;
            if enable_rollout {
                while !state.is_terminal() && depth < 200 {
                    // Horizon Check
                    if horizon == SearchHorizon::TurnEnd && state.turn > start_turn {
                        break;
                    }

                    state.generate_legal_actions(db, &mut self.legal_buffer);
                    if self.legal_buffer.is_empty() { break; }

                    let chunk_action = *self.legal_buffer.choose(&mut self.rng).unwrap();
                    let _ = state.step(db, chunk_action);
                    depth += 1;
                }
            }
            profiler.simulation += t_simulation.elapsed();

            // 5. Backpropagation
            let t_backprop = std::time::Instant::now();
            let reward_p0 = eval_fn(&state, db);

            let mut curr = Some(node_idx);
            while let Some(idx) = curr {
                let node_p_moved = self.nodes[idx].player_just_moved;
                let node = &mut self.nodes[idx];
                node.visit_count += 1;

                if node_p_moved == 0 {
                    node.value_sum += reward_p0;
                } else {
                    node.value_sum += 1.0 - reward_p0;
                }
                curr = node.parent;
            }
            profiler.backpropagation += t_backprop.elapsed();
        }

        let mut stats: Vec<(i32, f32, u32)> = self.nodes[0].children.iter()
            .map(|&(act, idx)| {
                let child = &self.nodes[idx];
                (act, child.value_sum / child.visit_count as f32, child.visit_count)
            })
            .collect();

        stats.sort_by_key(|&(_, _, v)| std::cmp::Reverse(v));
        (stats, profiler)
    }

    fn select_child(nodes: &[Node], node_idx: usize) -> usize {
        let node = &nodes[node_idx];
        let mut best_score = f32::NEG_INFINITY;
        let mut best_child = 0;

        let log_n = (node.visit_count as f32).ln();

        for &(_, child_idx) in &node.children {
            let child = &nodes[child_idx];
            // UCB1
            let exploit = child.value_sum / child.visit_count as f32;
            let explore = 1.4 * (log_n / child.visit_count as f32).sqrt();
            let score = exploit + explore;

            if score > best_score {
                best_score = score;
                best_child = child_idx;
            }
        }
        best_child
    }

    fn heuristic_eval(state: &GameState, db: &CardDatabase, p0_baseline: u32, p1_baseline: u32, eval_mode: EvalMode, p0_deck_stats: Option<([f32; 7], f32)>, p1_deck_stats: Option<([f32; 7], f32)>) -> f32 {
        let score0 = Self::evaluate_player(state, db, 0, p0_baseline, p0_deck_stats);

        if eval_mode == EvalMode::Solitaire {
             // Solitaire: Only care about P0 score. Normalize to [0,1].
             // Max possible score estimate: 3 lives * 5 + 2 bonus + 5 board + 2 hand = ~25
             return (score0 / 25.0).clamp(0.0, 1.0);
        }

        let score1 = Self::evaluate_player(state, db, 1, p1_baseline, p1_deck_stats);

        let mut final_val = (score0 - score1) * 0.5 + 0.5;

        // "Win the Live" Tie-breaker (Volume Lead)
        // If we are in Performance/LiveResult, checking volume matters for resolving ties.
        // Assuming equal lives, higher volume is better.
        let p0_vol = state.players[0].current_turn_volume;
        let p1_vol = state.players[1].current_turn_volume;

        if p0_vol > p1_vol {
            final_val += 0.05;
        } else if p1_vol > p0_vol {
            final_val -= 0.05;
        }

        final_val.clamp(0.0, 1.0)
    }

    fn evaluate_player(state: &GameState, db: &CardDatabase, p_idx: usize, baseline_score: u32, deck_stats: Option<([f32; 7], f32)>) -> f32 {
        let p = &state.players[p_idx];
        let mut score = 0.0;

        // 1. Success Lives (The Goal)
        // Heavily weight cleared lives.
        score += p.success_lives.len() as f32 * 5.0; // Increased weight to prioritize winning

        // Bonus for score increase relative to baseline
        if p.success_lives.len() > baseline_score as usize {
            score += 2.0;
        }

        // 2. Board State (Stage)
        let mut stage_hearts = [0u32; 7];
        let mut stage_blades = 0;

        for i in 0..3 {
            let h = state.get_effective_hearts(p_idx, i, db);
            for color in 0..7 { stage_hearts[color] += h[color] as u32; }
            stage_blades += state.get_effective_blades(p_idx, i, db);
        }
        // Small bonus for board presence/power
        score += stage_blades as f32 * 0.1;

        // 3. Deck Expectations (Yells)
        // Calculate expected value from deck based on probability
        let (avg_hearts, avg_vol) = if let Some(stats) = deck_stats {
            stats
        } else {
            Self::calculate_deck_expectations(&p.deck, db)
        };
        let expected_yell_hearts: Vec<f32> = avg_hearts.iter().map(|&h| h * stage_blades as f32).collect();
        let expected_volume = avg_vol * stage_blades as f32;

        // 4. Live Clearing Probability
        let mut max_prob = 0.0;
        for &cid in &p.live_zone {
            if cid >= 0 {
                if let Some(l) = db.get_live(cid as u16) {
                    let prob = Self::calculate_live_success_prob(
                        l,
                        &stage_hearts,
                        &expected_yell_hearts,
                        &p.heart_req_reductions
                    );
                    // Reward probability of clearing.
                    score += prob * (l.score as f32 * 2.0); // Higher weight for potential clears
                    if prob > max_prob { max_prob = prob; }
                }
            }
        }

        // Reward Volume if we are likely to clear at least one live
        if max_prob > 0.5 {
            score += (expected_volume + p.current_turn_volume as f32) * max_prob * 0.5;
        }

        // 5. Hand Quality (Construction)
        let hand_val = Self::calculate_hand_quality(state, db, p_idx);
        score += hand_val * 0.15;

        // 6. Resources & Discard
        score += p.hand.len() as f32 * 0.05;
        // Energy efficiency (using available energy)
        let unused_energy = p.tapped_energy.iter().filter(|&&t| !t).count();
        score += unused_energy as f32 * 0.01;

        // Discard Recovery Potential
        // If hand has recovery, value good stuff in discard
        let has_recovery = p.hand.iter().any(|&cid| {
            if let Some(m) = db.get_member(cid) {
                m.abilities.iter().any(|a| Self::has_opcode(&a.bytecode, 15) || Self::has_opcode(&a.bytecode, 17))
            } else { false }
        });

        if has_recovery {
            let discard_val = p.discard.iter().filter(|&&cid| {
                db.get_live(cid).is_some() || db.get_member(cid).map_or(false, |m| m.cost >= 3)
            }).count();
            score += discard_val as f32 * 0.1;
        }

        score
    }

    fn calculate_deck_expectations(deck: &[u16], db: &CardDatabase) -> ([f32; 7], f32) {
        if deck.is_empty() { return ([0.0; 7], 0.0); }

        // Since deck is determinized (shuffled) in the leaf node state,
        // we could just look at the top N cards if we knew N (blades).
        // But `blades` changes. So getting average stats of the *remaining* deck is more robust.

        let mut total_hearts = [0.0; 7];
        let mut total_vol = 0.0;
        let count = deck.len() as f32;

        for &cid in deck {
             if let Some(m) = db.get_member(cid) {
                 for i in 0..7 { total_hearts[i] += m.blade_hearts[i] as f32; }
                 total_vol += m.volume_icons as f32;
             } else if let Some(l) = db.get_live(cid) {
                 for i in 0..7 { total_hearts[i] += l.blade_hearts[i] as f32; }
                 total_vol += l.volume_icons as f32;
             }
        }

        let avg_hearts = total_hearts.map(|v| v / count);
        let avg_vol = total_vol / count;
        (avg_hearts, avg_vol)
    }

    fn calculate_live_success_prob(live: &LiveCard, stage_hearts: &[u32; 7], expected_yell_hearts: &[f32], reductions: &[i32; 7]) -> f32 {
        let mut prob;

        // 1. Hearts Check
        let mut needed = live.required_hearts;
        // Apply reductions
        for i in 0..7 {
            needed[i] = (needed[i] as i32 - reductions[i]).max(0) as u8;
        }

        let mut satisfied = 0.0;
        let mut total_req = 0.0;
        let mut wildcards_avail = stage_hearts[6] as f32 + expected_yell_hearts[6];

        // Specific Colors
        for i in 0..6 {
            let req = needed[i] as f32;
            total_req += req;
            let have = stage_hearts[i] as f32 + expected_yell_hearts[i];

            if have >= req {
                satisfied += req;
            } else {
                satisfied += have;
                let deficit = req - have;
                let used_wild = wildcards_avail.min(deficit);
                satisfied += used_wild;
                wildcards_avail -= used_wild;
            }
        }

        // ANY (Star) Requirement
        let any_req = needed[6] as f32;
        total_req += any_req;

        let used_wild = wildcards_avail.min(any_req);
        satisfied += used_wild;
        let mut remaining_any = any_req - used_wild;

        if remaining_any > 0.0 {
            // Use surpluses
            for i in 0..6 {
                let req = needed[i] as f32;
                let have = stage_hearts[i] as f32 + expected_yell_hearts[i];
                let surplus = (have - req).max(0.0);
                let used = surplus.min(remaining_any);
                satisfied += used;
                remaining_any -= used;
                if remaining_any <= 0.0 { break; }
            }
        }

        if total_req > 0.0 {
            prob = (satisfied / total_req).clamp(0.0, 1.0);

            // Non-linear scaling: getting 90% is much better than 50%
            prob = prob.powf(0.5); // Square root makes it optimistic? No, we want to penalize failure.
            // Actually, close to 1.0 is good.
            if prob >= 1.0 { prob = 1.2; } // Guaranteed win bonus
        } else {
            prob = 1.2; // Free live
        }

        prob
    }

    fn calculate_hand_quality(state: &GameState, db: &CardDatabase, p_idx: usize) -> f32 {
        let p = &state.players[p_idx];
        let mut val = 0.0;

        // In Mulligan phase, assume we will have ~3 energy soon for cost evaluations
        let is_mulligan = match state.phase {
            Phase::MulliganP1 | Phase::MulliganP2 => true,
            _ => false,
        };
        let max_energy = if is_mulligan { 3 } else { p.energy_zone.len() as u32 };

        for (i, &cid) in p.hand.iter().enumerate() {
            let card_val = Self::calculate_card_potential(cid, db, max_energy);

            // If card is selected for mulligan, it's effectively "gone" but replaced by an "average" card
            // Average card potential is roughly 1.0
            if is_mulligan && ((p.mulligan_selection >> i) & 1u64 == 1) {
                val += 1.0; // Expected potential of replacement
            } else {
                val += card_val;
            }
        }
        val
    }

    fn calculate_card_potential(cid: u16, db: &CardDatabase, max_energy: u32) -> f32 {
        if let Some(m) = db.get_member(cid) {
            let mut score = 0.0;
            // Cost Efficiency
            let stat_sum: u32 = m.hearts.iter().map(|&x| x as u32).sum();
            score += (m.blades as f32 + stat_sum as f32) / (m.cost as f32 + 1.0);

            // Penalty for unplayable high-cost cards
            if m.cost > max_energy {
                let diff = m.cost - max_energy;
                score -= diff as f32 * 0.5;
            }

            // Fast Flag Checks (O(1))
            let f = m.ability_flags;

            if (f & FLAG_DRAW) != 0 { score += 0.5; }
            if (f & FLAG_SEARCH) != 0 { score += 0.6; }
            if (f & FLAG_RECOVER) != 0 { score += 0.4; }
            if (f & FLAG_BUFF) != 0 { score += 0.3; }
            if (f & FLAG_CHARGE) != 0 { score += 0.8; }
            if (f & FLAG_TEMPO) != 0 { score += 0.2; }
            if (f & FLAG_REDUCE) != 0 { score += 0.4; }
            if (f & FLAG_BOOST) != 0 { score += 0.5; }
            if (f & FLAG_TRANSFORM) != 0 { score += 0.3; }
            if (f & FLAG_WIN_COND) != 0 { score += 0.6; }

            score
        } else if let Some(l) = db.get_live(cid) {
            // Lives in hand are good if we can clear them, but bad if they clog
            // Simple heuristic: Score value
            l.score as f32 * 0.2
        } else {
            0.0
        }
    }

    fn has_opcode(bytecode: &[i32], target_op: i32) -> bool {
        let mut i = 0;
        while i < bytecode.len() {
            if i + 3 >= bytecode.len() { break; }
            let op = bytecode[i];
            if op == target_op { return true; }
            i += 4;
        }
        false
    }
}

#[cfg(feature = "nn")]
pub struct HybridMCTS {
    pub session: Arc<Mutex<Session>>,
    pub neural_weight: f32,
    pub skip_rollout: bool,
    pub rng: SmallRng,
}

#[cfg(feature = "nn")]
impl HybridMCTS {
    pub fn new(session: Arc<Mutex<Session>>, neural_weight: f32, skip_rollout: bool) -> Self {
        Self {
            session,
            neural_weight,
            skip_rollout,
            rng: SmallRng::from_os_rng(),
        }
    }

    pub fn get_suggestions(&mut self, state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32) -> Vec<(i32, f32, u32)> {
        let start = std::time::Instant::now();
        let (stats, profile) = self.search(state, db, num_sims, timeout_sec);
        let duration = start.elapsed();
        if num_sims > 10 {
            profile.print(duration);
        }
        stats
    }

    pub fn search(&mut self, root_state: &GameState, db: &CardDatabase, num_sims: usize, timeout_sec: f32) -> (Vec<(i32, f32, u32)>, MCTSProfiler) {
        let session_arc = self.session.clone();
        let neural_weight = self.neural_weight;
        let mut mcts = MCTS::new();

        mcts.run_mcts_config(root_state, db, num_sims, timeout_sec, SearchHorizon::GameEnd, false, !self.skip_rollout, |state: &GameState, db: &CardDatabase| {
            if state.is_terminal() {
                return match state.get_winner() {
                    0 => 1.0,
                    1 => 0.0,
                    _ => 0.5,
                };
            }

            // Normal Heuristic Baseline
            let h_val = MCTS::heuristic_eval(state, db, root_state.players[0].score, root_state.players[1].score, EvalMode::Normal, None, None);

            // NN Evaluation
            let input_vec = state.encode_state(db);
            let mut session = session_arc.lock().unwrap();

            let input_shape = [1, input_vec.len()];

            // Try to create input tensor and run using (shape, vec) which is version-agnostic
            if let Ok(input_tensor) = ort::value::Value::from_array((input_shape, input_vec)) {
                if let Ok(outputs) = session.run(ort::inputs![input_tensor]) {
                    // Try to get value output. AlphaNet has 'output_1' or index 1
                    let val_opt = outputs.get("output_1")
                        .or_else(|| outputs.get("value"));

                    if let Some(v_val) = val_opt {
                        if let Ok((_, v_slice)) = v_val.try_extract_tensor::<f32>() {
                            let nn_val = v_slice[0];
                            let nn_norm = (nn_val * 0.5 + 0.5).clamp(0.0, 1.0) as f32;
                            return h_val * (1.0 - neural_weight) + nn_norm * neural_weight;
                        }
                    } else if outputs.len() > 1 {
                        // Fallback to index if names missing
                        if let Ok((_, v_slice)) = outputs[1].try_extract_tensor::<f32>() {
                            let nn_val = v_slice[0];
                            let nn_norm = (nn_val * 0.5 + 0.5).clamp(0.0, 1.0) as f32;
                            return h_val * (1.0 - neural_weight) + nn_norm * neural_weight;
                        }
                    }
                }
            }

            h_val
        })
    }
}