diff --git a/rust_hft/prediction-markets/config/research_missions/polymarket-btc-5m.example.json b/rust_hft/prediction-markets/config/research_missions/polymarket-btc-5m.example.json
index fa4d741dd..2496d6e62 100644
--- a/rust_hft/prediction-markets/config/research_missions/polymarket-btc-5m.example.json
+++ b/rust_hft/prediction-markets/config/research_missions/polymarket-btc-5m.example.json
@@ -15,7 +15,7 @@
   "horizon": "5m",
   "time_cohort_boundary_ms": 0,
   "prompt_snapshot_id": "sha256:2b55ba0e724dfc9f5a040911e397058a09c6229e42064e4998739d617b368dcb",
-  "search_policy_snapshot_id": "sha256:f8d80d3125cba1d7c2e4e23e68f3ab63314533ec8df2912a5cf8bff3a60dc352",
+  "search_policy_snapshot_id": "sha256:3375b087cdaf642916b8a4a99a0816a96200c2860fa042046a12b6741f59bbd3",
   "search_budget": {
     "max_candidates": 6,
     "max_llm_calls": 2,
diff --git a/rust_hft/prediction-markets/config/research_missions/polymarket-sol-5m.example.json b/rust_hft/prediction-markets/config/research_missions/polymarket-sol-5m.example.json
index 2c5556d57..18786cf8e 100644
--- a/rust_hft/prediction-markets/config/research_missions/polymarket-sol-5m.example.json
+++ b/rust_hft/prediction-markets/config/research_missions/polymarket-sol-5m.example.json
@@ -15,7 +15,7 @@
   "horizon": "5m",
   "time_cohort_boundary_ms": 0,
   "prompt_snapshot_id": "sha256:0816ebccf4c75ee6bdcfe315b253c84ddf1808a38687ebe3b924b87bd72a52a9",
-  "search_policy_snapshot_id": "sha256:f8d80d3125cba1d7c2e4e23e68f3ab63314533ec8df2912a5cf8bff3a60dc352",
+  "search_policy_snapshot_id": "sha256:3375b087cdaf642916b8a4a99a0816a96200c2860fa042046a12b6741f59bbd3",
   "search_budget": {
     "max_candidates": 6,
     "max_llm_calls": 2,
diff --git a/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts.rs b/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts.rs
index 0ba00abc6..53dca10ec 100644
--- a/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts.rs
+++ b/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts.rs
@@ -401,11 +401,14 @@ enum ExpansionAction {
 }
 
 fn deterministic_actions(profile: SettlementProbabilityComponentProfile) -> Vec<ExpansionAction> {
-    eligible_components(profile)
-        .iter()
-        .copied()
-        .map(|component| ExpansionAction::Increase { component })
-        .collect()
+    match profile {
+        SettlementProbabilityComponentProfile::FullSurface => eligible_components(profile)
+            .iter()
+            .copied()
+            .map(|component| ExpansionAction::Increase { component })
+            .collect(),
+        SettlementProbabilityComponentProfile::MarketMidpointOnly => Vec::new(),
+    }
 }
 
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
@@ -518,6 +521,22 @@ impl PredictionMctsCheckpoint {
         }) {
             return Err("prediction MCTS baseline identity mismatch".to_string());
         }
+        let market_midpoint_only = self.config.component_profile
+            == SettlementProbabilityComponentProfile::MarketMidpointOnly;
+        if market_midpoint_only && !self.config.llm_advisor_sha256.is_empty() {
+            return Err("market-midpoint-only search does not accept LLM advice".to_string());
+        }
+        if market_midpoint_only
+            && (self.proposed > 1
+                || self.nodes.len() != 1
+                || self.seen_blends.len() != 1
+                || !self.nodes[0].children.is_empty()
+                || !self.nodes[0].unexpanded_actions.is_empty())
+        {
+            return Err(
+                "market-midpoint-only search evaluates only its canonical baseline".to_string(),
+            );
+        }
         validate_tree(&self.nodes, 0, self.config.max_depth)
             .map_err(|error| format!("invalid prediction MCTS tree: {error}"))?;
         for (node_id, node) in self.nodes.iter().enumerate() {
@@ -580,7 +599,16 @@ impl PredictionMctsCheckpoint {
             let ordinal = candidate_id
                 .strip_prefix(&prefix)
                 .and_then(|value| value.parse::<usize>().ok());
-            if *node_id == 0
+            let canonical_baseline_pending = *node_id == 0
+                && self.config.component_profile
+                    == SettlementProbabilityComponentProfile::MarketMidpointOnly
+                && self.pending.len() == 1
+                && self.proposed == 1
+                && ordinal == Some(0)
+                && node.visits == 0
+                && node.total_reward == 0.0
+                && node.best_reward.is_none();
+            if (*node_id == 0 && !canonical_baseline_pending)
                 || !pending_nodes.insert(*node_id)
                 || ordinal.is_none_or(|ordinal| {
                     ordinal >= self.proposed || !pending_ordinals.insert(ordinal)
@@ -768,6 +796,11 @@ impl PredictionMctsEngine {
         if mission.search_budget.max_candidates == 0 {
             return Err("prediction MCTS requires a non-zero candidate budget".to_string());
         }
+        if component_profile == SettlementProbabilityComponentProfile::MarketMidpointOnly
+            && !llm_advice.is_empty()
+        {
+            return Err("market-midpoint-only search does not accept LLM advice".to_string());
+        }
         if !llm_advice.is_empty() && mission.search_budget.max_llm_calls == 0 {
             return Err("LLM advice requires a non-zero governed LLM-call budget".to_string());
         }
@@ -843,6 +876,16 @@ impl PredictionMctsEngine {
         if self.checkpoint.proposed >= self.checkpoint.config.max_candidates {
             return Err("prediction MCTS candidate budget is exhausted".to_string());
         }
+        if self.checkpoint.config.component_profile
+            == SettlementProbabilityComponentProfile::MarketMidpointOnly
+        {
+            if self.checkpoint.proposed == 0 {
+                return Ok(self.mark_candidate_pending(0));
+            }
+            return Err(
+                "market-midpoint-only search evaluates only its canonical baseline".to_string(),
+            );
+        }
         for _ in 0..256 {
             let parent = select_expandable(
                 &self.checkpoint.nodes,
@@ -856,31 +899,40 @@ impl PredictionMctsEngine {
             if !self.checkpoint.seen_blends.insert(blend_key(&blend)) {
                 continue;
             }
-            let probability_blend_sha256 = blend_digest(&blend);
-            let candidate_id = format!(
-                "{}{}",
-                candidate_id_prefix(&self.checkpoint.config, &blend),
-                self.checkpoint.proposed
-            );
-            self.checkpoint.proposed += 1;
-            self.checkpoint
-                .pending
-                .insert(candidate_id.clone(), node_id);
-            return Ok(PredictionMctsCandidate {
-                candidate_id,
-                identity: self.checkpoint.config.identity.clone(),
-                probability_blend_sha256,
-                source: self.checkpoint.nodes[node_id].source,
-                probability_blend: blend,
-            });
+            return Ok(self.mark_candidate_pending(node_id));
         }
         Err("prediction MCTS could not produce a novel candidate".to_string())
     }
 
+    fn mark_candidate_pending(&mut self, node_id: usize) -> PredictionMctsCandidate {
+        let blend = self.checkpoint.nodes[node_id].blend.clone();
+        let candidate_id = format!(
+            "{}{}",
+            candidate_id_prefix(&self.checkpoint.config, &blend),
+            self.checkpoint.proposed
+        );
+        self.checkpoint.proposed += 1;
+        self.checkpoint
+            .pending
+            .insert(candidate_id.clone(), node_id);
+        PredictionMctsCandidate {
+            candidate_id,
+            identity: self.checkpoint.config.identity.clone(),
+            probability_blend_sha256: blend_digest(&blend),
+            source: self.checkpoint.nodes[node_id].source,
+            probability_blend: blend,
+        }
+    }
+
     pub(crate) fn has_expandable_candidate(&self) -> Result<bool, String> {
         if self.checkpoint.proposed >= self.checkpoint.config.max_candidates {
             return Ok(false);
         }
+        if self.checkpoint.config.component_profile
+            == SettlementProbabilityComponentProfile::MarketMidpointOnly
+        {
+            return Ok(self.checkpoint.proposed == 0);
+        }
         Ok(select_expandable(
             &self.checkpoint.nodes,
             0,
@@ -1170,7 +1222,7 @@ mod tests {
     }
 
     #[test]
-    fn market_midpoint_profile_never_expands_an_unsupported_component() {
+    fn market_midpoint_profile_evaluates_only_its_canonical_baseline() {
         let mut mission = mission();
         mission.search_budget.max_candidates = 3;
         mission.search_budget.max_llm_calls = 1;
@@ -1186,32 +1238,62 @@ mod tests {
         )
         .expect("market-midpoint-only engine");
 
-        for _ in 0..3 {
-            let candidate = engine.propose().expect("eligible deterministic candidate");
-            assert_eq!(candidate.probability_blend.chainlink_digital_weight, 0.0);
-            assert_eq!(candidate.probability_blend.distance_lob_vol_weight, 0.0);
-            assert_eq!(candidate.probability_blend.event_surface_weight, 0.0);
-            assert_eq!(candidate.probability_blend.existing_model_weight, 0.0);
-            let evaluation = FakeEvaluator
-                .evaluate_training(&candidate)
-                .expect("training evidence");
-            engine
-                .observe(&candidate.candidate_id, &evaluation)
-                .expect("observe eligible candidate");
-        }
+        let candidate = engine.propose().expect("canonical baseline candidate");
+        assert_eq!(candidate.source, PredictionExpansionSource::Baseline);
+        assert_eq!(
+            candidate.probability_blend,
+            market_midpoint_blend("baseline").into()
+        );
+        engine.checkpoint().expect("durable pending baseline");
+        let evaluation = FakeEvaluator
+            .evaluate_training(&candidate)
+            .expect("training evidence");
+        engine
+            .observe(&candidate.candidate_id, &evaluation)
+            .expect("observe canonical baseline");
+        assert!(!engine.has_expandable_candidate().unwrap());
+        let before = engine.checkpoint().expect("observed baseline checkpoint");
+        let error = engine
+            .propose()
+            .expect_err("market-midpoint-only search must stop after its baseline");
+        assert!(error.contains("canonical baseline"), "{error}");
+        assert_eq!(engine.checkpoint().unwrap(), before);
+
+        let mut forged = before.clone();
+        forged.proposed = 2;
+        let error = engine
+            .restore_checkpoint(forged)
+            .expect_err("expanded midpoint-only checkpoint must fail closed");
+        assert!(error.contains("canonical baseline"), "{error}");
+        assert_eq!(engine.checkpoint().unwrap(), before);
+
+        let advisor = validate_single_proposal(market_midpoint_blend("advisor")).unwrap();
+        let mut forged = before.clone();
+        forged
+            .config
+            .llm_advisor_sha256
+            .push(blend_digest(&advisor));
+        forged.nodes[0]
+            .unexpanded_actions
+            .push(ExpansionAction::LlmAdvisor { blend: advisor });
+        let error = engine
+            .restore_checkpoint(forged)
+            .expect_err("midpoint-only checkpoint must reject LLM authority");
+        assert!(error.contains("does not accept LLM advice"), "{error}");
+        assert_eq!(engine.checkpoint().unwrap(), before);
 
         let error = PredictionMctsEngine::new_with_component_profile(
             &mission,
             market_midpoint_blend("baseline"),
-            vec![blend("invalid_advisor")],
+            vec![market_midpoint_blend("advisor")],
             7,
             1.4,
             3,
             SettlementProbabilityComponentProfile::MarketMidpointOnly,
         )
         .err()
-        .expect("unsupported advisor must fail");
-        assert!(error.contains("ineligible"));
+        .expect("market-midpoint-only profile must reject LLM advice");
+        assert!(error.contains("does not accept LLM advice"));
     }
 
     #[test]
diff --git a/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts_run.rs b/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts_run.rs
index d812fa7a6..ad91dcefb 100644
--- a/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts_run.rs
+++ b/rust_hft/prediction-markets/crates/ploy-research/src/prediction_mcts_run.rs
@@ -1377,7 +1377,7 @@ mod tests {
     }
 
     #[test]
-    fn reduced_profile_selects_when_the_tree_exhausts_before_the_budget() {
+    fn reduced_profile_selects_its_only_canonical_candidate() {
         let output = temp_dir("reduced-profile-tree-exhaustion");
         let mut client = FakeClient { calls: 0 };
         let mut evaluator = FakeEvaluator::default();
@@ -1393,7 +1393,8 @@ mod tests {
         .expect("tree exhaustion selects the best evaluated candidate");
 
         assert_eq!(evaluator.calls.last().map(String::as_str), Some("held_out"));
-        assert_eq!(evaluator.candidate_ids.len(), 4);
+        assert_eq!(evaluator.candidate_ids.len(), 2);
+        assert_eq!(evaluator.candidate_ids[0], evaluator.candidate_ids[1]);
     }
 
     #[test]