diff --git a/ml/src/ensemble/coordinator.rs b/ml/src/ensemble/coordinator.rs
index 796cbd537..83b53a1ff 100644
--- a/ml/src/ensemble/coordinator.rs
+++ b/ml/src/ensemble/coordinator.rs
@@ -4,9 +4,13 @@
 //! from multiple ML models (DQN, PPO, TFT, MAMBA-2, Liquid, TLOB) for production trading decisions.
 //! Supports dynamic weighting based on performance and model diversity metrics.
 
+use crate::ensemble::conviction_gates::{
+    ConvictionGateConfig, ConvictionGateEvaluator, ConvictionGateOutcome, GateInput, TradingSession,
+};
 use crate::ensemble::inference_adapter::{FeatureVector, ModelInferenceAdapter};
 use crate::ensemble::{EnsembleDecision, ModelVote, ModelWeight, TradingAction};
 use crate::{Features, MLError, MLResult, ModelPrediction};
+use chrono::{Timelike, Utc};
 use std::collections::HashMap;
 use std::sync::Arc;
 use tokio::sync::RwLock;
@@ -34,6 +38,9 @@ pub struct EnsembleCoordinator {
 
     /// Real model inference adapters for production predictions
     adapters: Vec<Box<dyn ModelInferenceAdapter>>,
+
+    /// Conviction gate evaluator (optional — None means no gating)
+    conviction_gates: Option<ConvictionGateEvaluator>,
 }
 
 impl std::fmt::Debug for EnsembleCoordinator {
@@ -77,9 +84,28 @@ impl EnsembleCoordinator {
             model_weights: Arc::new(RwLock::new(HashMap::new())),
             config,
             adapters: Vec::new(),
+            conviction_gates: Some(ConvictionGateEvaluator::new(ConvictionGateConfig {
+                // Allow all sessions by default — trading service narrows to Regular
+                allowed_sessions: vec![
+                    TradingSession::PreMarket,
+                    TradingSession::Regular,
+                    TradingSession::AfterHours,
+                ],
+                ..ConvictionGateConfig::default()
+            })),
         }
     }
 
+    /// Replace conviction gate configuration
+    pub fn set_conviction_gates(&mut self, config: ConvictionGateConfig) {
+        self.conviction_gates = Some(ConvictionGateEvaluator::new(config));
+    }
+
+    /// Get mutable reference to conviction gate evaluator
+    pub fn conviction_gates_mut(&mut self) -> Option<&mut ConvictionGateEvaluator> {
+        self.conviction_gates.as_mut()
+    }
+
     /// Add a real model inference adapter to the ensemble.
     /// When an adapter's model_name matches a registered model, its predictions
     /// are used for ensemble inference.
@@ -129,6 +155,50 @@ impl EnsembleCoordinator {
             .aggregate(predictions, &*self.model_weights.read().await)
             .await?;
 
+        // Apply conviction gates if configured
+        if let Some(ref gates) = self.conviction_gates {
+            let gate_input = GateInput {
+                confidence: decision.confidence,
+                disagreement_rate: decision.disagreement_rate,
+                quorum_ratio: Self::calculate_quorum_ratio(&decision),
+                healthy_models: self.adapters.iter().filter(|a| a.is_ready()).count(),
+                total_models: self.adapters.len(),
+                current_session: Self::current_trading_session(),
+                regime_volatility: Self::extract_regime_volatility(features),
+            };
+
+            match gates.evaluate(&gate_input) {
+                ConvictionGateOutcome::Passed(pass_result) => {
+                    info!(
+                        "Conviction gates passed: score={:.3}, {} gates evaluated",
+                        pass_result.conviction_score,
+                        pass_result.gate_details.len()
+                    );
+                    let mut gated_decision = decision;
+                    gated_decision.metadata.insert(
+                        "conviction_score".into(),
+                        serde_json::Value::from(pass_result.conviction_score),
+                    );
+                    return Ok(gated_decision);
+                }
+                ConvictionGateOutcome::Rejected(rejection) => {
+                    info!("Conviction gate rejected: {:?} — forcing HOLD", rejection);
+                    let mut hold_decision = EnsembleDecision::new(
+                        TradingAction::Hold,
+                        decision.confidence,
+                        0.0,
+                        decision.disagreement_rate,
+                        decision.model_votes,
+                    );
+                    hold_decision.metadata.insert(
+                        "gate_rejection".into(),
+                        serde_json::Value::String(format!("{:?}", rejection)),
+                    );
+                    return Ok(hold_decision);
+                }
+            }
+        }
+
         info!(
             "Ensemble decision: {:?}, confidence: {:.3}, disagreement: {:.3}",
             decision.action, decision.confidence, decision.disagreement_rate
@@ -292,6 +362,46 @@ impl EnsembleCoordinator {
 
         Ok(())
     }
+
+    /// Calculate quorum ratio (fraction of models agreeing on majority direction)
+    fn calculate_quorum_ratio(decision: &EnsembleDecision) -> f64 {
+        if decision.model_votes.is_empty() {
+            return 0.0;
+        }
+        let majority_action = &decision.action;
+        let agreeing = decision
+            .model_votes
+            .values()
+            .filter(|v| TradingAction::from_signal(v.signal, 0.3) == *majority_action)
+            .count();
+        agreeing as f64 / decision.model_votes.len() as f64
+    }
+
+    /// Determine current trading session (Eastern Time, simplified UTC-5)
+    fn current_trading_session() -> TradingSession {
+        let now = Utc::now();
+        let et_hour = (now.hour() + 24 - 5) % 24;
+        let et_minute = now.minute();
+        let et_time = et_hour * 60 + et_minute;
+
+        match et_time {
+            t if t < 240 => TradingSession::AfterHours, // 00:00-04:00 ET
+            t if t < 570 => TradingSession::PreMarket,  // 04:00-09:30 ET
+            t if t < 960 => TradingSession::Regular,     // 09:30-16:00 ET
+            _ => TradingSession::AfterHours,             // 16:00-24:00 ET
+        }
+    }
+
+    /// Extract regime volatility from feature vector
+    /// Regime features are at indices 48-50 in the 51-dim standard feature vector
+    fn extract_regime_volatility(features: &Features) -> f64 {
+        features
+            .values
+            .get(48)
+            .copied()
+            .unwrap_or(0.01)
+            .abs()
+    }
 }
 
 impl Default for EnsembleCoordinator {
diff --git a/ml/src/ensemble/mod.rs b/ml/src/ensemble/mod.rs
index 50b023d78..15a699dcc 100644
--- a/ml/src/ensemble/mod.rs
+++ b/ml/src/ensemble/mod.rs
@@ -50,8 +50,8 @@ pub use metrics::{
 pub use training_integration::EnsembleTrainingIntegration;
 pub use inference_adapter::{EnsemblePrediction, FeatureVector, ModelInferenceAdapter, PredictionMeta};
 pub use conviction_gates::{
-    ConvictionGateConfig, ConvictionGateOutcome, GateEvaluation, GatePassResult, GateRejection,
-    TradingSession,
+    ConvictionGateConfig, ConvictionGateEvaluator, ConvictionGateOutcome, GateEvaluation,
+    GateInput, GatePassResult, GateRejection, TradingSession,
 };
 
 /// Errors that can occur in ensemble operations