diff --git a/crates/applications/multi_core_inverter/src/main.rs b/crates/applications/multi_core_inverter/src/main.rs
index 01dc641..ed74ab5 100644
--- a/crates/applications/multi_core_inverter/src/main.rs
+++ b/crates/applications/multi_core_inverter/src/main.rs
@@ -43,7 +43,7 @@ use coremem::meas;
 use coremem::real::{self, Real as _};
 use coremem::sim::spirv::{self, SpirvSim};
 use coremem::sim::units::Seconds;
-use coremem::stim::{CurlVectorField, Exp, Gated, ModulatedVectorField, Shifted, StimExt as _, StimuliVec};
+use coremem::stim::{CurlVectorField, Exp, Gated, ModulatedVectorField, Shifted, StimExt as _};
 use coremem::Driver;
 
 type R = real::R32;
@@ -248,7 +248,6 @@ fn main() {
         })
         .collect();
     assert_eq!(stim.len(), 5);
-    let stim = StimuliVec::from_vec(stim);
 
 
     let wire_mat = IsomorphicConductor::new(1e6f32.cast::<R>());
@@ -304,7 +303,10 @@ fn main() {
     let duration = Seconds(params.clock_phase_duration * (num_cycles + 3) as f32);
 
     // let stim = DynStimuli::from_vec(stim.map(MapIntoBoxStimulus).into_vec());
-    let mut driver = driver.with_stimulus(stim);
+    let mut driver = driver.with_concrete_stimulus();
+    for s in stim {
+        driver.add_stimulus(s);
+    }
 
     let prefix = "out/applications/multi_core_inverter/20-4ns-1e7A/";
     let _ = std::fs::create_dir_all(&prefix);
diff --git a/crates/coremem/src/driver.rs b/crates/coremem/src/driver.rs
index 5b582d5..833d398 100644
--- a/crates/coremem/src/driver.rs
+++ b/crates/coremem/src/driver.rs
@@ -6,8 +6,9 @@ use crate::real::Real;
 use crate::render::{self, MultiRenderer, Renderer};
 use crate::sim::AbstractSim;
 use crate::sim::units::{Frame, Time};
-use crate::stim::{Stimulus, DynStimuli};
+use crate::stim::{DynStimuli, Stimulus, StimuliVec};
 use coremem_cross::compound::list;
+use coremem_cross::step::SimMeta;
 
 use log::{info, trace};
 use serde::{Deserialize, Serialize};
@@ -17,7 +18,7 @@ use std::sync::mpsc::{sync_channel, SyncSender, Receiver};
 use std::time::Instant;
 use threadpool::ThreadPool;
 
-pub struct Driver<S, Stim=DynStimuli> {
+pub struct Driver<S, Stim=DriverStimulusDynVec> {
     state: S,
     renderer: Arc<MultiRenderer<S>>,
     // TODO: use Rayon's thread pool?
@@ -56,26 +57,32 @@ impl<S: AbstractSim, Stim> Driver<S, Stim> {
     pub fn add_measurement<Meas: AbstractMeasurement<S> + 'static>(&mut self, m: Meas) {
         self.measurements.push(Arc::new(m));
     }
+
+    pub fn add_stimulus<SNew>(&mut self, s: SNew)
+        where Stim: Pushable<SNew>
+    {
+        self.stimuli.push(s)
+    }
 }
 
-impl<S: AbstractSim> Driver<S, DynStimuli> {
+impl<S: AbstractSim> Driver<S, DriverStimulusDynVec> {
     pub fn new(state: S) -> Self {
-        Self::new_with_stim(state, DynStimuli::default())
-    }
-    pub fn add_stimulus<Stim: Stimulus + 'static>(&mut self, s: Stim) {
-        self.stimuli.push(Box::new(s))
+        Self::new_with_stim(state, DriverStimulusDynVec::default())
     }
 }
 
 impl<S: AbstractSim> Driver<S, list::Empty> {
-    pub fn new_unboxed_stim(state: S) -> Self {
+    pub fn new_list_stim(state: S) -> Self {
         Self::new_with_stim(state, list::Empty::default())
     }
 }
 
 impl<S: AbstractSim, Stim> Driver<S, Stim> {
-    pub fn with_add_stimulus<E>(self, s: E) -> Driver<S, list::Prepended<E, Stim>>
-        where Stim: list::Prependable
+    /// add a stimulus onto a list of non-monomorphized stimuli.
+    /// this necessarily must return a new Self.
+    /// (well, with enough tuning we could actually Box just the first reference...
+    pub fn with_add_stimulus<E>(self, s: E) -> Driver<S, list::Appended<Stim, E>>
+        where Stim: list::Appendable<E>
     {
         Driver {
             state: self.state,
@@ -83,7 +90,7 @@ impl<S: AbstractSim, Stim> Driver<S, Stim> {
             render_pool: self.render_pool,
             render_channel: self.render_channel,
             measurements: self.measurements,
-            stimuli: self.stimuli.prepend(s),
+            stimuli: self.stimuli.append(s),
             sim_end_time: self.sim_end_time,
             diag: self.diag,
             last_diag_time: self.last_diag_time,
@@ -102,6 +109,9 @@ impl<S: AbstractSim, Stim> Driver<S, Stim> {
             last_diag_time: self.last_diag_time,
         }
     }
+    pub fn with_concrete_stimulus<T>(self) -> Driver<S, DriverStimulusVec<T>> {
+        self.with_stimulus(DriverStimulusVec::<T>::default())
+    }
 }
 
 impl<S: AbstractSim, Stim> Driver<S, Stim> {
@@ -199,7 +209,7 @@ where
 impl<S, Stim> Driver<S, Stim>
 where
     S: AbstractSim + Clone + Default + Send + 'static,
-    Stim: Stimulus,
+    Stim: DriverStimulus,
 {
     fn render(&mut self) {
         self.diag.instrument_render_prep(|| {
@@ -238,7 +248,8 @@ where
         }
         trace!("step begin");
         self.diag.instrument_step(can_step as u64, || {
-            self.state.step_multiple(can_step, &self.stimuli);
+            let stim = self.stimuli.optimized_for(self.state.meta());
+            self.state.step_multiple(can_step, stim.as_ref());
         });
         trace!("step end");
         if self.last_diag_time.elapsed().as_secs_f64() >= 5.0 {
@@ -290,3 +301,70 @@ where
         self.sim_end_time = None;
     }
 }
+
+pub enum ValueOrRef<'a, T> {
+    Value(T),
+    Ref(&'a T),
+}
+impl<'a, T> AsRef<T> for ValueOrRef<'a, T> {
+    fn as_ref(&self) -> &T {
+        match self {
+            ValueOrRef::Value(x) => &x,
+            ValueOrRef::Ref(x) => x,
+        }
+    }
+}
+
+/// gives an opportunity to optimize a Stimulus for a specific setting
+/// before passing it off to the simulation.
+pub trait DriverStimulus {
+    type Optimized: Stimulus;
+    fn optimized_for<'a>(
+        &'a self, meta: SimMeta<f32>
+    ) -> ValueOrRef<'a, Self::Optimized>;
+}
+
+pub trait Pushable<T> {
+    fn push(&mut self, t: T);
+}
+
+pub struct DriverStimulusVec<T>(StimuliVec<T>);
+
+impl<T> Default for DriverStimulusVec<T> {
+    fn default() -> Self {
+        Self(Default::default())
+    }
+}
+
+impl<T: Stimulus> DriverStimulus for DriverStimulusVec<T> {
+    type Optimized = StimuliVec<T>;
+    fn optimized_for<'a>(
+        &'a self, _meta: SimMeta<f32>
+    ) -> ValueOrRef<'a, Self::Optimized> {
+        ValueOrRef::Ref(&self.0)
+    }
+}
+
+impl<T> Pushable<T> for DriverStimulusVec<T> {
+    fn push(&mut self, t: T) {
+        self.0.push(t)
+    }
+}
+
+#[derive(Default)]
+pub struct DriverStimulusDynVec(DynStimuli);
+
+impl DriverStimulus for DriverStimulusDynVec {
+    type Optimized = DynStimuli;
+    fn optimized_for<'a>(
+        &'a self, _meta: SimMeta<f32>
+    ) -> ValueOrRef<'a, Self::Optimized> {
+        ValueOrRef::Ref(&self.0)
+    }
+}
+
+impl<T: Stimulus + 'static> Pushable<T> for DriverStimulusDynVec {
+    fn push(&mut self, t: T) {
+        self.0.push(Box::new(t))
+    }
+}