diff --git a/Cargo.lock b/Cargo.lock
index 326b2b189..8d5b20d95 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -849,6 +849,19 @@ dependencies = [
  "spider-tdl",
 ]
 
+[[package]]
+name = "huntsman-nn-core"
+version = "0.1.0"
+
+[[package]]
+name = "huntsman-nn-tasks"
+version = "0.1.0"
+dependencies = [
+ "huntsman-nn-core",
+ "serde",
+ "spider-tdl",
+]
+
 [[package]]
 name = "hyper"
 version = "1.10.1"
diff --git a/Cargo.toml b/Cargo.toml
index c624e3e7e..288f2a808 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -14,6 +14,8 @@ members = [
   "components/spider-utils",
   "examples/huntsman/complex/tasks",
   "examples/huntsman/complex/types",
+  "examples/huntsman/nn/core",
+  "examples/huntsman/nn/tasks",
   "tests/huntsman/em-runtime",
   "tests/huntsman/integration-test-tasks",
   "tests/huntsman/task-executor",
diff --git a/examples/huntsman/nn/core/Cargo.toml b/examples/huntsman/nn/core/Cargo.toml
new file mode 100644
index 000000000..790f5fa46
--- /dev/null
+++ b/examples/huntsman/nn/core/Cargo.toml
@@ -0,0 +1,9 @@
+[package]
+name = "huntsman-nn-core"
+version = "0.1.0"
+edition = "2024"
+publish = false
+
+[lib]
+name = "huntsman_nn_core"
+path = "src/lib.rs"
diff --git a/examples/huntsman/nn/core/src/lib.rs b/examples/huntsman/nn/core/src/lib.rs
new file mode 100644
index 000000000..efc7800f6
--- /dev/null
+++ b/examples/huntsman/nn/core/src/lib.rs
@@ -0,0 +1,160 @@
+//! Pure neuron math for the Spider end-to-end neural-network test workload.
+//!
+//! A dense-layer neuron computes `activation(weighted_sum(inputs) + bias)` over a fixed fan-in of
+//! 25 scalar `double` inputs.
+
+/// The fixed neuron fan-in: each neuron consumes exactly this many scalar inputs.
+pub const NUM_INPUTS: usize = 25;
+
+/// The fixed per-input weights, one per input position. Deterministic values calculated as
+/// (`WEIGHTS[k] = (k + 1) * 0.01 * (-1)^k`), alternating in sign starting positive.
+pub const WEIGHTS: [f64; NUM_INPUTS] = [
+    0.01, -0.02, 0.03, -0.04, 0.05, -0.06, 0.07, -0.08, 0.09, -0.10, 0.11, -0.12, 0.13, -0.14,
+    0.15, -0.16, 0.17, -0.18, 0.19, -0.20, 0.21, -0.22, 0.23, -0.24, 0.25,
+];
+
+/// The fixed bias added to the weighted sum before the activation.
+pub const BIAS: f64 = 0.5;
+
+/// # Returns
+///
+/// The rectified-linear activation `max(0.0, x)`.
+#[must_use]
+pub const fn relu(x: f64) -> f64 {
+    f64::max(0.0, x)
+}
+
+/// # Returns
+///
+/// The logistic sigmoid activation `1.0 / (1.0 + exp(-x))`.
+#[must_use]
+pub fn sigmoid(x: f64) -> f64 {
+    1.0 / (1.0 + f64::exp(-x))
+}
+
+/// # Returns
+///
+/// The identity activation `x`.
+#[must_use]
+pub const fn identity(x: f64) -> f64 {
+    x
+}
+
+/// # Returns
+///
+/// The rectified-linear activation of the weighted sum of `inputs` plus [`BIAS`].
+#[must_use]
+pub fn dense_relu(inputs: &[f64; NUM_INPUTS]) -> f64 {
+    relu(weighted_sum(inputs))
+}
+
+/// # Returns
+///
+/// The logistic sigmoid of the weighted sum of `inputs` plus [`BIAS`].
+#[must_use]
+pub fn dense_sigmoid(inputs: &[f64; NUM_INPUTS]) -> f64 {
+    sigmoid(weighted_sum(inputs))
+}
+
+/// # Returns
+///
+/// The weighted sum of `inputs` plus [`BIAS`], unchanged by the activation.
+#[must_use]
+pub fn dense_identity(inputs: &[f64; NUM_INPUTS]) -> f64 {
+    identity(weighted_sum(inputs))
+}
+
+/// # Returns
+///
+/// The weighted sum `sum(WEIGHTS[k] * inputs[k]) + BIAS`.
+fn weighted_sum(inputs: &[f64; NUM_INPUTS]) -> f64 {
+    let mut acc = BIAS;
+    for (w, x) in WEIGHTS.iter().zip(inputs.iter()) {
+        acc += w * x;
+    }
+    acc
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// Relative-tolerance float equality used to compare hand-computed and computed values.
+    fn assert_approx_eq(actual: f64, expected: f64) {
+        let diff = (actual - expected).abs();
+        let tol = 1.0e-12_f64 * (1.0 + expected.abs());
+        assert!(
+            diff <= tol,
+            "actual={actual}, expected={expected}, diff={diff}, tol={tol}",
+        );
+    }
+
+    #[test]
+    fn test_relu() {
+        assert_approx_eq(relu(-1.0), 0.0);
+        assert_approx_eq(relu(0.0), 0.0);
+        assert_approx_eq(relu(2.5), 2.5);
+    }
+
+    #[test]
+    fn test_sigmoid() {
+        assert_approx_eq(sigmoid(0.0), 0.5);
+        assert_approx_eq(sigmoid(100.0), 1.0);
+        assert_approx_eq(sigmoid(-100.0), 0.0);
+        assert!(sigmoid(-1.0) < sigmoid(0.0));
+        assert!(sigmoid(0.0) < sigmoid(1.0));
+    }
+
+    #[test]
+    fn test_identity() {
+        assert_approx_eq(identity(-3.0), -3.0);
+        assert_approx_eq(identity(0.0), 0.0);
+        assert_approx_eq(identity(7.25), 7.25);
+    }
+
+    #[test]
+    fn test_weighted_sum_all_zero_inputs_equals_bias() {
+        let inputs = [0.0_f64; NUM_INPUTS];
+        assert_approx_eq(weighted_sum(&inputs), BIAS);
+    }
+
+    #[test]
+    fn test_weighted_sum_all_one_inputs() {
+        let inputs = [1.0_f64; NUM_INPUTS];
+        assert_approx_eq(weighted_sum(&inputs), 0.63);
+    }
+
+    #[test]
+    fn test_dense_relu() {
+        let zero = [0.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_relu(&zero), 0.5);
+
+        let ones = [1.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_relu(&ones), 0.63);
+
+        // Negative weighted sum (large negative inputs) clamps to 0 under relu.
+        let neg = [-1000.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_relu(&neg), 0.0);
+    }
+
+    #[test]
+    fn test_dense_sigmoid() {
+        let zero = [0.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_sigmoid(&zero), sigmoid(BIAS));
+
+        let ones = [1.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_sigmoid(&ones), sigmoid(0.63));
+    }
+
+    #[test]
+    fn test_dense_identity() {
+        let zero = [0.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_identity(&zero), 0.5);
+
+        let ones = [1.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_identity(&ones), 0.63);
+
+        let neg = [-1000.0_f64; NUM_INPUTS];
+        assert_approx_eq(dense_identity(&neg), weighted_sum(&neg));
+    }
+}
diff --git a/examples/huntsman/nn/tasks/Cargo.toml b/examples/huntsman/nn/tasks/Cargo.toml
new file mode 100644
index 000000000..29dee028d
--- /dev/null
+++ b/examples/huntsman/nn/tasks/Cargo.toml
@@ -0,0 +1,18 @@
+[package]
+name = "huntsman-nn-tasks"
+version = "0.1.0"
+edition = "2024"
+publish = false
+
+[lib]
+crate-type = ["cdylib"]
+name = "nn"
+path = "src/lib.rs"
+
+[dependencies]
+huntsman-nn-core = { path = "../core" }
+serde = { version = "1.0.228", features = ["derive"] }
+spider-tdl = {
+  path = "../../../../components/spider-tdl",
+  features = ["derive"]
+}
diff --git a/examples/huntsman/nn/tasks/src/lib.rs b/examples/huntsman/nn/tasks/src/lib.rs
new file mode 100644
index 000000000..d398dd67a
--- /dev/null
+++ b/examples/huntsman/nn/tasks/src/lib.rs
@@ -0,0 +1,124 @@
+//! Reference TDL package: dense-neuron computation.
+
+#![allow(clippy::too_many_arguments)]
+
+mod task_decl {
+    use spider_tdl::TaskContext;
+    use spider_tdl::TdlError;
+    use spider_tdl::r#std::double;
+    use spider_tdl::task;
+
+    #[task(name = "neuron::dense_relu")]
+    pub fn dense_relu(
+        _ctx: TaskContext,
+        x0: double,
+        x1: double,
+        x2: double,
+        x3: double,
+        x4: double,
+        x5: double,
+        x6: double,
+        x7: double,
+        x8: double,
+        x9: double,
+        x10: double,
+        x11: double,
+        x12: double,
+        x13: double,
+        x14: double,
+        x15: double,
+        x16: double,
+        x17: double,
+        x18: double,
+        x19: double,
+        x20: double,
+        x21: double,
+        x22: double,
+        x23: double,
+        x24: double,
+    ) -> Result<double, TdlError> {
+        Ok(huntsman_nn_core::dense_relu(&[
+            x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, x16, x17, x18,
+            x19, x20, x21, x22, x23, x24,
+        ]))
+    }
+
+    #[task(name = "neuron::dense_sigmoid")]
+    pub fn dense_sigmoid(
+        _ctx: TaskContext,
+        x0: double,
+        x1: double,
+        x2: double,
+        x3: double,
+        x4: double,
+        x5: double,
+        x6: double,
+        x7: double,
+        x8: double,
+        x9: double,
+        x10: double,
+        x11: double,
+        x12: double,
+        x13: double,
+        x14: double,
+        x15: double,
+        x16: double,
+        x17: double,
+        x18: double,
+        x19: double,
+        x20: double,
+        x21: double,
+        x22: double,
+        x23: double,
+        x24: double,
+    ) -> Result<double, TdlError> {
+        Ok(huntsman_nn_core::dense_sigmoid(&[
+            x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, x16, x17, x18,
+            x19, x20, x21, x22, x23, x24,
+        ]))
+    }
+
+    #[task(name = "neuron::dense_identity")]
+    pub fn dense_identity(
+        _ctx: TaskContext,
+        x0: double,
+        x1: double,
+        x2: double,
+        x3: double,
+        x4: double,
+        x5: double,
+        x6: double,
+        x7: double,
+        x8: double,
+        x9: double,
+        x10: double,
+        x11: double,
+        x12: double,
+        x13: double,
+        x14: double,
+        x15: double,
+        x16: double,
+        x17: double,
+        x18: double,
+        x19: double,
+        x20: double,
+        x21: double,
+        x22: double,
+        x23: double,
+        x24: double,
+    ) -> Result<double, TdlError> {
+        Ok(huntsman_nn_core::dense_identity(&[
+            x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15, x16, x17, x18,
+            x19, x20, x21, x22, x23, x24,
+        ]))
+    }
+}
+
+spider_tdl::register_tdl_package! {
+    package_name: "nn",
+    tasks: [
+        task_decl::dense_relu,
+        task_decl::dense_sigmoid,
+        task_decl::dense_identity,
+    ],
+}