diff --git a/src/gradient_solver/backprop.rs b/src/gradient_solver/backprop.rs
index 9f5f0ed..d3e2c34 100644
--- a/src/gradient_solver/backprop.rs
+++ b/src/gradient_solver/backprop.rs
@@ -2,7 +2,7 @@ use num::ToPrimitive;
 
 use crate::{
     derivable::NeuraLoss, layer::NeuraTrainableLayerBackprop, layer::NeuraTrainableLayerSelf,
-    network::NeuraTrainableNetworkBase,
+    network::NeuraOldTrainableNetworkBase,
 };
 
 use super::*;
@@ -20,12 +20,12 @@ impl<Loss> NeuraBackprop<Loss> {
 impl<
         Input,
         Target,
-        Trainable: NeuraTrainableNetworkBase<Input>,
+        Trainable: NeuraOldTrainableNetworkBase<Input>,
         Loss: NeuraLoss<Trainable::Output, Target = Target> + Clone,
     > NeuraGradientSolver<Input, Target, Trainable> for NeuraBackprop<Loss>
 where
     <Loss as NeuraLoss<Trainable::Output>>::Output: ToPrimitive,
-    Trainable: for<'a> NeuraTrainableNetwork<Input, (&'a NeuraBackprop<Loss>, &'a Target)>,
+    Trainable: for<'a> NeuraOldTrainableNetwork<Input, (&'a NeuraBackprop<Loss>, &'a Target)>,
 {
     fn get_gradient(
         &self,
diff --git a/src/gradient_solver/forward_forward.rs b/src/gradient_solver/forward_forward.rs
index 01ce34d..c68f713 100644
--- a/src/gradient_solver/forward_forward.rs
+++ b/src/gradient_solver/forward_forward.rs
@@ -30,7 +30,7 @@ impl<
         F,
         Act: Clone + NeuraDerivable<f64>,
         Input,
-        Trainable: NeuraTrainableNetwork<Input, NeuraForwardPair<Act>, Output = DVector<F>>,
+        Trainable: NeuraOldTrainableNetwork<Input, NeuraForwardPair<Act>, Output = DVector<F>>,
     > NeuraGradientSolver<Input, bool, Trainable> for NeuraForwardForward<Act>
 where
     F: ToPrimitive,
@@ -138,7 +138,7 @@ impl<
     fn map_epsilon<From, To, Gradient, Cb: Fn(From) -> To>(
         &self,
         rec_opt_output: Self::Output<From, Gradient>,
-        callback: Cb
+        _callback: Cb
     ) -> Self::Output<To, Gradient> {
         rec_opt_output
     }
diff --git a/src/gradient_solver/mod.rs b/src/gradient_solver/mod.rs
index d5497d2..275c964 100644
--- a/src/gradient_solver/mod.rs
+++ b/src/gradient_solver/mod.rs
@@ -6,7 +6,7 @@ pub use forward_forward::NeuraForwardForward;
 
 use crate::{
     layer::NeuraTrainableLayerBase,
-    network::{NeuraTrainableNetwork, NeuraTrainableNetworkBase},
+    network::{NeuraOldTrainableNetwork, NeuraOldTrainableNetworkBase},
 };
 
 pub trait NeuraGradientSolverBase {
@@ -37,7 +37,7 @@ pub trait NeuraGradientSolverTransient<Input, Layer: NeuraTrainableLayerBase<Inp
     ) -> Self::Output<To, Gradient>;
 }
 
-pub trait NeuraGradientSolver<Input, Target, Trainable: NeuraTrainableNetworkBase<Input>> {
+pub trait NeuraGradientSolver<Input, Target, Trainable: NeuraOldTrainableNetworkBase<Input>> {
     fn get_gradient(
         &self,
         trainable: &Trainable,
diff --git a/src/network/mod.rs b/src/network/mod.rs
index e13de8f..abc148a 100644
--- a/src/network/mod.rs
+++ b/src/network/mod.rs
@@ -2,11 +2,11 @@ use crate::{
     algebra::NeuraVectorSpace, gradient_solver::{NeuraGradientSolverBase, NeuraGradientSolverFinal}, layer::NeuraLayer,
 };
 
-pub mod residual;
+// pub mod residual;
 pub mod sequential;
 
 // TODO: extract regularize from this, so that we can drop the trait constraints on NeuraSequential's impl
-pub trait NeuraTrainableNetworkBase<Input>: NeuraLayer<Input> {
+pub trait NeuraOldTrainableNetworkBase<Input>: NeuraLayer<Input> {
     type Gradient: NeuraVectorSpace;
     type LayerOutput;
 
@@ -21,7 +21,7 @@ pub trait NeuraTrainableNetworkBase<Input>: NeuraLayer<Input> {
     fn prepare(&mut self, train_iteration: bool);
 }
 
-pub trait NeuraTrainableNetwork<Input, Optimizer>: NeuraTrainableNetworkBase<Input>
+pub trait NeuraOldTrainableNetwork<Input, Optimizer>: NeuraOldTrainableNetworkBase<Input>
 where
     Optimizer: NeuraGradientSolverBase,
 {
@@ -33,7 +33,7 @@ where
 }
 
 impl<Input: Clone, Optimizer: NeuraGradientSolverFinal<Input>>
-    NeuraTrainableNetwork<Input, Optimizer> for ()
+    NeuraOldTrainableNetwork<Input, Optimizer> for ()
 {
     fn traverse(
         &self,
diff --git a/src/network/residual/layer_impl.rs b/src/network/residual/layer_impl.rs
index 4e14d1b..b5c67cd 100644
--- a/src/network/residual/layer_impl.rs
+++ b/src/network/residual/layer_impl.rs
@@ -174,16 +174,25 @@ where
         let child_result = self.child_network.traverse(&rest, optimizer);
         // TODO: maybe move this to a custom impl of NeuraGradientSolverTransient for NeuraResidualInput?
         // Or have a different set of traits for NeuraTrainableNetwork specific to NeuraResidualNodes
-        let child_result = optimizer.map_epsilon(child_result, |epsilon| {
+        let child_result = optimizer.map_epsilon(child_result, |_epsilon| {
             // Pop the first value from `epsilon`, then:
             // - compute its sum
             // - use it to compute the outcoming epsilon of the current layer
             // - split the oucoming epsilon into its original components, and push those back onto the rest
             // At this point, the value for `epsilon` in the gradient solver's state should be ready for another iteration,
             // with the first value containing the unsummed incoming epsilon values from the downstream layers
-            todo!();
+            todo!()
         });
 
+        optimizer.eval_layer(
+            &self.layer,
+            &layer_input,
+            &layer_output,
+            &layer_intermediary,
+            child_result,
+            |this_gradient, child_gradient| (this_gradient, Box::new(child_gradient))
+        );
+
         todo!();
     }
 }
diff --git a/src/network/sequential/mod.rs b/src/network/sequential/mod.rs
index a917a29..c2fbc63 100644
--- a/src/network/sequential/mod.rs
+++ b/src/network/sequential/mod.rs
@@ -1,6 +1,6 @@
-use super::{NeuraTrainableNetwork, NeuraTrainableNetworkBase};
+use super::{NeuraOldTrainableNetwork, NeuraOldTrainableNetworkBase};
 use crate::{
-    gradient_solver::{NeuraGradientSolverFinal, NeuraGradientSolverTransient},
+    gradient_solver::{NeuraGradientSolverTransient},
     layer::{
         NeuraLayer, NeuraPartialLayer, NeuraShape, NeuraTrainableLayerBase, NeuraTrainableLayerSelf,
     },
@@ -82,8 +82,8 @@ impl<Layer, ChildNetwork> NeuraSequential<Layer, ChildNetwork> {
 impl<
         Input,
         Layer: NeuraTrainableLayerBase<Input> + NeuraTrainableLayerSelf<Input>,
-        ChildNetwork: NeuraTrainableNetworkBase<Layer::Output>,
-    > NeuraTrainableNetworkBase<Input> for NeuraSequential<Layer, ChildNetwork>
+        ChildNetwork: NeuraOldTrainableNetworkBase<Layer::Output>,
+    > NeuraOldTrainableNetworkBase<Input> for NeuraSequential<Layer, ChildNetwork>
 {
     type Gradient = (Layer::Gradient, Box<ChildNetwork::Gradient>);
     type LayerOutput = Layer::Output;
@@ -114,7 +114,7 @@ impl<
 }
 
 /// A dummy implementation of `NeuraTrainableNetwork`, which simply calls `loss.eval` in `backpropagate`.
-impl<Input: Clone> NeuraTrainableNetworkBase<Input> for () {
+impl<Input: Clone> NeuraOldTrainableNetworkBase<Input> for () {
     type Gradient = ();
     type LayerOutput = Input;
 
@@ -143,10 +143,10 @@ impl<
         Input,
         Layer: NeuraTrainableLayerBase<Input> + NeuraTrainableLayerSelf<Input>,
         Optimizer: NeuraGradientSolverTransient<Input, Layer>,
-        ChildNetwork: NeuraTrainableNetworkBase<Layer::Output>,
-    > NeuraTrainableNetwork<Input, Optimizer> for NeuraSequential<Layer, ChildNetwork>
+        ChildNetwork: NeuraOldTrainableNetworkBase<Layer::Output>,
+    > NeuraOldTrainableNetwork<Input, Optimizer> for NeuraSequential<Layer, ChildNetwork>
 where
-    ChildNetwork: NeuraTrainableNetwork<Layer::Output, Optimizer>,
+    ChildNetwork: NeuraOldTrainableNetwork<Layer::Output, Optimizer>,
 {
     fn traverse(
         &self,
diff --git a/src/train.rs b/src/train.rs
index 24e1ce4..648d693 100644
--- a/src/train.rs
+++ b/src/train.rs
@@ -1,6 +1,6 @@
 use crate::{
     algebra::NeuraVectorSpace, gradient_solver::NeuraGradientSolver,
-    network::NeuraTrainableNetworkBase,
+    network::NeuraOldTrainableNetworkBase,
 };
 
 #[non_exhaustive]
@@ -73,7 +73,7 @@ impl NeuraBatchedTrainer {
     pub fn train<
         Input: Clone,
         Target: Clone,
-        Network: NeuraTrainableNetworkBase<Input>,
+        Network: NeuraOldTrainableNetworkBase<Input>,
         GradientSolver: NeuraGradientSolver<Input, Target, Network>,
         Inputs: IntoIterator<Item = (Input, Target)>,
     >(
@@ -84,7 +84,7 @@ impl NeuraBatchedTrainer {
         test_inputs: &[(Input, Target)],
     ) -> Vec<(f64, f64)>
     where
-        <Network as NeuraTrainableNetworkBase<Input>>::Gradient: std::fmt::Debug,
+        <Network as NeuraOldTrainableNetworkBase<Input>>::Gradient: std::fmt::Debug,
     {
         let mut losses = Vec::new();
         let mut iter = inputs.into_iter();