diff --git a/crates/asap-aware-mapping/src/pass/major.rs b/crates/asap-aware-mapping/src/pass/major.rs
index 6deffdfcd..7956ea094 100644
--- a/crates/asap-aware-mapping/src/pass/major.rs
+++ b/crates/asap-aware-mapping/src/pass/major.rs
@@ -70,7 +70,7 @@ impl OptimizationPass for MajorPass {
             demand,
             lifecycle.now_ms,
             lifecycle.horizon,
-            lifecycle.capabilities,
+            models.capabilities,
             models.cost,
         )
         .map_err(OptimizeError::LifecycleSelection)?;
@@ -132,7 +132,7 @@ impl OptimizationPass for MajorPass {
                     },
                     lifecycle.now_ms,
                     lifecycle.horizon,
-                    lifecycle.capabilities,
+                    models.capabilities,
                     models.cost,
                 )
                 .map_err(|source| OptimizeError::LifecycleAssembly {
@@ -171,7 +171,7 @@ impl OptimizationPass for MajorPass {
                 },
                 lifecycle.now_ms,
                 lifecycle.horizon,
-                lifecycle.capabilities,
+                models.capabilities,
                 models.cost,
             )
             .map_err(|source| OptimizeError::LifecycleAssembly {
diff --git a/crates/asap-aware-mapping/src/pass/mod.rs b/crates/asap-aware-mapping/src/pass/mod.rs
index f9dd0ac2d..34e691cc0 100644
--- a/crates/asap-aware-mapping/src/pass/mod.rs
+++ b/crates/asap-aware-mapping/src/pass/mod.rs
@@ -48,6 +48,7 @@ pub struct PlanningModels<'a> {
     pub cost: &'a dyn CostModel,
     pub accuracy: &'a dyn AccuracyModel,
     pub evidence: &'a dyn AccuracyEvidenceProvider,
+    pub capabilities: SummaryMaintenanceLifecycleCapabilities,
 }
 
 impl<'a> PlanningModels<'a> {
@@ -60,6 +61,7 @@ impl<'a> PlanningModels<'a> {
             cost,
             accuracy,
             evidence,
+            capabilities: SummaryMaintenanceLifecycleCapabilities::ALL,
         }
     }
 
@@ -71,6 +73,7 @@ impl<'a> PlanningModels<'a> {
             cost: &DEFAULT_COST_MODEL,
             accuracy: &DEFAULT_ACCURACY_MODEL,
             evidence: &NO_ACCURACY_EVIDENCE,
+            capabilities: SummaryMaintenanceLifecycleCapabilities::ALL,
         }
     }
 
@@ -88,6 +91,14 @@ impl<'a> PlanningModels<'a> {
         self.evidence = evidence;
         self
     }
+
+    pub fn with_capabilities(
+        mut self,
+        capabilities: SummaryMaintenanceLifecycleCapabilities,
+    ) -> Self {
+        self.capabilities = capabilities;
+        self
+    }
 }
 
 /// Supplying this asks the pass to also decide summary maintenance versus raw
@@ -99,15 +110,13 @@ pub struct LifecycleInput {
     pub now_ms: u64,
     /// Seconds. Required to turn recurring demand into a finite total.
     pub horizon: Option<Horizon>,
-    pub capabilities: SummaryMaintenanceLifecycleCapabilities,
 }
 
 impl LifecycleInput {
-    pub fn new(now_ms: u64, capabilities: SummaryMaintenanceLifecycleCapabilities) -> Self {
+    pub fn new(now_ms: u64) -> Self {
         Self {
             now_ms,
             horizon: None,
-            capabilities,
         }
     }
 
@@ -123,7 +132,8 @@ pub struct OptimizationInput<'a> {
     pub workload: &'a ParsedWorkload,
     pub models: PlanningModels<'a>,
     /// Every plan carries the maintenance-versus-recomputation decision, so
-    /// the planning clock and runtime capabilities are always required.
+    /// the planning clock is always required. Runtime capabilities are in
+    /// `models` with the deployment's cost and accuracy models.
     pub lifecycle: LifecycleInput,
 }
 
diff --git a/crates/integration-tests/tests/operator_design_examples.rs b/crates/integration-tests/tests/operator_design_examples.rs
index ace32b1c4..4800ad528 100644
--- a/crates/integration-tests/tests/operator_design_examples.rs
+++ b/crates/integration-tests/tests/operator_design_examples.rs
@@ -261,8 +261,7 @@ async fn batch_planning_replaces_and_shares_summary_operators() {
     use asap_aware_mapping::cost_model::{Cost, DefaultCostModel};
     use asap_aware_mapping::pass::PlanningModels;
     use asap_aware_mapping::{
-        CostModel, CostRate, LifecycleInput, SummaryMaintenanceLifecycleCapabilities,
-        SummaryMaintenanceLifecycleCostInputs,
+        CostModel, CostRate, LifecycleInput, SummaryMaintenanceLifecycleCostInputs,
     };
     use asap_physical_operators::{
         physical_planner::{compile, InputContract},
@@ -336,7 +335,7 @@ async fn batch_planning_replaces_and_shares_summary_operators() {
         &workload,
         FrontendInput::Sql { catalog: &catalog },
         PlanningModels::builtin().with_cost(&Costs),
-        LifecycleInput::new(0, SummaryMaintenanceLifecycleCapabilities::default()),
+        LifecycleInput::new(0),
     ))
     .await
     .unwrap();
diff --git a/crates/planner/src/lib.rs b/crates/planner/src/lib.rs
index 4f527f148..501bc9ead 100644
--- a/crates/planner/src/lib.rs
+++ b/crates/planner/src/lib.rs
@@ -53,8 +53,8 @@ pub struct UserInput<'a> {
     pub workload: &'a PlanningWorkload,
     pub frontend_specific: FrontendInput<'a>,
     pub models: PlanningModels<'a>,
-    /// Planning clock and runtime capabilities for the
-    /// maintenance-versus-recomputation decision every plan carries.
+    /// Planning clock for the maintenance-versus-recomputation decision every
+    /// plan carries. Runtime capabilities are supplied with `models`.
     pub lifecycle: LifecycleInput,
     /// `None` uses [`MajorPass`]. A black-box caller never sets this.
     pub pass: Option<&'a dyn OptimizationPass>,
diff --git a/crates/planner/tests/e2e_plan.rs b/crates/planner/tests/e2e_plan.rs
index 3ab55b101..20d83d4e7 100644
--- a/crates/planner/tests/e2e_plan.rs
+++ b/crates/planner/tests/e2e_plan.rs
@@ -7,9 +7,7 @@ use asap_aware_mapping::pass::{
     OptimizationInput, OptimizationPass, OptimizeError, PlanOutput, PlanningModels,
 };
 use asap_aware_mapping::replacement::default_strategies_with_evidence;
-use asap_aware_mapping::{
-    search_workload_with_targets, Horizon, LifecycleInput, SummaryMaintenanceLifecycleCapabilities,
-};
+use asap_aware_mapping::{search_workload_with_targets, Horizon, LifecycleInput};
 use asap_frontend_sql::{lower_sql_dialect, SqlCatalog};
 use asap_planner::{e2e_plan, FrontendInput, PlanError, UserInput, UserInputError};
 use asap_types::pre_asap::schema::{DataType, Field, Schema};
@@ -43,7 +41,7 @@ fn batch(sql: &str) -> BatchEntry {
 /// The planning clock and default capabilities, no horizon: the least a
 /// caller can supply.
 fn lifecycle() -> LifecycleInput {
-    LifecycleInput::new(NOW_MS, SummaryMaintenanceLifecycleCapabilities::default())
+    LifecycleInput::new(NOW_MS)
 }
 
 fn lineitem_catalog() -> SqlCatalog {
@@ -165,6 +163,49 @@ async fn builtin_cost_model_cannot_price_lifecycles_and_falls_back_to_raw_recomp
     }
 }
 
+/// Runtime capabilities are deployment inputs alongside the cost and
+/// accuracy models, while LifecycleInput carries only the planning clock.
+#[tokio::test]
+async fn deployment_inputs_control_lifecycle_capabilities() {
+    let workload = sql_workload(
+        vec![batch("SELECT COUNT(DISTINCT l_orderkey) FROM lineitem")],
+        None,
+    );
+    let catalog = lineitem_catalog();
+    let models = PlanningModels::builtin().with_capabilities(
+        asap_aware_mapping::SummaryMaintenanceLifecycleCapabilities {
+            supports_ephemeral: true,
+            supports_prepared: false,
+            supports_shared: false,
+            supports_continuously_maintained: false,
+        },
+    );
+
+    let output = e2e_plan(UserInput::new(
+        &workload,
+        FrontendInput::Sql { catalog: &catalog },
+        models,
+        lifecycle(),
+    ))
+    .await
+    .expect("workload remains plannable with restricted capabilities");
+
+    assert!(output
+        .plans
+        .iter()
+        .flat_map(|plan| &plan.plan.deployments)
+        .all(
+            |deployment| deployment.alternatives.iter().all(|alternative| {
+                !matches!(
+                    alternative.summary_maintenance_lifecycle,
+                    asap_types::post_asap::SummaryMaintenanceLifecycle::Prepared { .. }
+                        | asap_types::post_asap::SummaryMaintenanceLifecycle::Shared { .. }
+                        | asap_types::post_asap::SummaryMaintenanceLifecycle::ContinuouslyMaintained
+                ) || alternative.rejection.is_some()
+            })
+        ));
+}
+
 /// A repeating SQL query reaches the optimizer. `lower_sql_batch` walks
 /// `query_batch` alone, so driving the frontend through it would drop exactly
 /// the entries whose recurrence the lifecycle stage reads.
@@ -338,10 +379,7 @@ fn rejects_disagreeing_planning_clocks() {
             histograms: None,
         },
         PlanningModels::builtin(),
-        LifecycleInput::new(
-            NOW_MS + 1,
-            SummaryMaintenanceLifecycleCapabilities::default(),
-        ),
+        LifecycleInput::new(NOW_MS + 1),
     );
 
     assert!(matches!(
diff --git a/crates/planner/tests/summary_sharing.rs b/crates/planner/tests/summary_sharing.rs
index 9b7a204ef..6cad6b8bf 100644
--- a/crates/planner/tests/summary_sharing.rs
+++ b/crates/planner/tests/summary_sharing.rs
@@ -10,7 +10,7 @@ use asap_aware_mapping::pass::{PlanOutput, PlanningModels};
 use asap_aware_mapping::replacement::{default_size_params, DEFAULT_DELTA};
 use asap_aware_mapping::{
     CostModel, CostRate, DefaultCostModel, Horizon, LifecycleInput, SummaryMaintenanceCapabilities,
-    SummaryMaintenanceLifecycleCapabilities, SummaryMaintenanceLifecycleCostInputs,
+    SummaryMaintenanceLifecycleCostInputs,
 };
 use asap_frontend_sql::SqlCatalog;
 use asap_planner::{e2e_plan, FrontendInput, UserInput};
@@ -103,8 +103,7 @@ fn repeating(query: &str, epsilon: f64) -> RepeatingEntry {
 }
 
 fn lifecycle() -> LifecycleInput {
-    LifecycleInput::new(NOW_MS, SummaryMaintenanceLifecycleCapabilities::default())
-        .with_horizon(Horizon(HORIZON_S))
+    LifecycleInput::new(NOW_MS).with_horizon(Horizon(HORIZON_S))
 }
 
 fn promql_workload(queries: &[(&str, f64)]) -> PlanningWorkload {