fix(ml): wire spread_cost_bps into RL training — commission + spread = total tx cost

Previously train_baseline_rl.rs only passed commission (tx_cost_bps) to
DQN/PPO trainers, ignoring bid-ask spread slippage. Now computes per-fold
average spread via spread_cost_bps() (same as evaluate_baseline) and passes
total cost (commission + spread) to both trainers.

Removes #[allow(dead_code)] — function is now used by all 4 example binaries.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

crates/ml/examples/baseline_common/mod.rs

@@ -193,7 +193,6 @@ pub fn list_subdirs(dir: &Path) -> Vec<String> {
 /// Uses the bid-ask spread model from `backtesting/src/slippage.rs`:
 /// half-spread = `tick_size` * `spread_ticks` / 2.  Round-trip pays the full spread.
 /// Converted to bps: `spread_price` / price * `10_000`.
-#[allow(dead_code)] // Used by evaluate_baseline, train_baseline_supervised — not all examples
 pub fn spread_cost_bps(price: f64, tick_size: f64, spread_ticks: f64) -> f64 {
     if price.abs() < 1e-10 {
         return 0.0;

crates/ml/examples/train_baseline_rl.rs

@@ -34,7 +34,7 @@ use ml::trainers::ppo::{PpoHyperparameters, PpoTrainer};
 #[allow(unreachable_pub)]
 mod baseline_common;
 use baseline_common::completion::{write_failure_marker, write_success_marker, CompletionMetrics};
-use baseline_common::load_all_bars;
+use baseline_common::{load_all_bars, spread_cost_bps};
 use common::metrics::{server as metrics_server, training_metrics as metrics};
 use ml::features::extraction::extract_ml_features;
 use ml::types::OHLCVBar;
@@ -292,9 +292,22 @@ fn train_dqn_fold(
     );
     info!("  [DQN] Fold {} -- {} train, {} val features", fold, train_features.len(), val_features.len());
 
+    // Compute average spread slippage in bps from training bar prices.
+    // Same pattern as evaluate_baseline: total_cost = commission + spread.
+    let avg_price = if train_bars.is_empty() {
+        0.0
+    } else {
+        train_bars.iter().map(|b| b.close).sum::<f64>() / train_bars.len() as f64
+    };
+    let avg_spread_bps = spread_cost_bps(avg_price, args.tick_size, args.spread_ticks);
+    let total_cost_bps = args.tx_cost_bps + avg_spread_bps;
+    info!("  [DQN] Fold {} total tx cost: {:.2} bps (commission {:.1} + spread {:.2})",
+        fold, total_cost_bps, args.tx_cost_bps, avg_spread_bps);
+
     // Build DQNHyperparameters -- hyperopt JSON overrides defaults, CLI args override both.
     // `..DQNHyperparameters::default()` enables all Rainbow components (PER, dueling,
     // distributional C51, noisy nets, gradient accumulation, mixed precision auto-detect).
+    // transaction_cost_multiplier scales internal reward cost by total trading friction.
     let hyperparams = DQNHyperparameters {
         learning_rate: hp_f64(hp, "learning_rate").unwrap_or(args.learning_rate),
         batch_size: hp_usize(hp, "batch_size").unwrap_or(args.batch_size),
@@ -310,6 +323,7 @@ fn train_dqn_fold(
         warmup_steps: 0,
         // Early stopping is managed by DQNTrainer internally
         early_stopping_enabled: true,
+        transaction_cost_multiplier: total_cost_bps,
         ..DQNHyperparameters::default()
     };
 
@@ -368,7 +382,7 @@ fn train_ppo_fold(
     fold: usize,
     train_features: &[[f64; 51]],
     _val_features: &[[f64; 51]],
-    _train_bars: &[OHLCVBar],
+    train_bars: &[OHLCVBar],
     _val_bars: &[OHLCVBar],
     args: &Args,
     output_dir: &Path,
@@ -382,6 +396,18 @@ fn train_ppo_fold(
         return Ok(f64::MAX);
     }
 
+    // Compute average spread slippage in bps from training bar prices.
+    // Same pattern as evaluate_baseline: total_cost = commission + spread.
+    let avg_price = if train_bars.is_empty() {
+        0.0
+    } else {
+        train_bars.iter().map(|b| b.close).sum::<f64>() / train_bars.len() as f64
+    };
+    let avg_spread_bps = spread_cost_bps(avg_price, args.tick_size, args.spread_ticks);
+    let total_cost_bps = args.tx_cost_bps + avg_spread_bps;
+    info!("  [PPO] Fold {} total tx cost: {:.2} bps (commission {:.1} + spread {:.2})",
+        fold, total_cost_bps, args.tx_cost_bps, avg_spread_bps);
+
     // Build PpoHyperparameters -- hyperopt JSON overrides defaults, CLI args override both.
     // Start from conservative() baseline so all fields have sane values.
     let hyperparams = PpoHyperparameters {
@@ -400,7 +426,7 @@ fn train_ppo_fold(
         minibatch_size: hp_usize(hp, "minibatch_size").unwrap_or(64),
         epochs: args.epochs,
         early_stopping_enabled: true,
-        transaction_cost_bps: args.tx_cost_bps / 100.0, // CLI is in bps (1.0 bps), trainer wants pct (0.01)
+        transaction_cost_bps: total_cost_bps / 100.0, // total (commission + spread) in bps → pct for trainer
         ..PpoHyperparameters::conservative()
     };