diff --git a/src/kernel.rs b/src/kernel.rs
index 3888343..6cf58e1 100644
--- a/src/kernel.rs
+++ b/src/kernel.rs
@@ -161,12 +161,6 @@ impl Kernel for Vec<Box<dyn Kernel>> {
         transition
     }
 
-    /*
-        def transition(self, t1, t2):
-        mats = [k.transition(t1, t2) for k in self.parts]
-        return block_diag(*mats)
-    */
-
     fn noise_cov(&self, t0: f64, t1: f64) -> Array2<f64> {
         let data = self
             .iter()
diff --git a/src/model.rs b/src/model.rs
index f3b0869..536e417 100644
--- a/src/model.rs
+++ b/src/model.rs
@@ -1,5 +1,7 @@
 mod binary;
+mod difference;
 mod ternary;
 
 pub use binary::*;
+pub use difference::*;
 pub use ternary::*;
diff --git a/src/model/difference.rs b/src/model/difference.rs
new file mode 100644
index 0000000..054a1e5
--- /dev/null
+++ b/src/model/difference.rs
@@ -0,0 +1,100 @@
+use std::f64;
+
+use crate::fitter::RecursiveFitter;
+use crate::item::Item;
+use crate::kernel::Kernel;
+use crate::observation::*;
+use crate::storage::Storage;
+
+#[derive(Clone, Copy)]
+pub enum DifferenceModelFitMethod {
+    Ep,
+    Kl,
+}
+
+pub struct DifferenceModel {
+    storage: Storage,
+    last_t: f64,
+    observations: Vec<GaussianObservation>,
+    last_method: Option<DifferenceModelFitMethod>,
+    var: f64,
+}
+
+impl DifferenceModel {
+    pub fn new(var: f64) -> Self {
+        DifferenceModel {
+            storage: Storage::new(),
+            last_t: f64::NEG_INFINITY,
+            observations: Vec::new(),
+            last_method: None,
+            var, // default = 1.0
+        }
+    }
+
+    pub fn add_item(&mut self, name: &str, kernel: Box<dyn Kernel>) {
+        if self.storage.contains_key(name) {
+            panic!("item '{}' already added", name);
+        }
+
+        self.storage.insert(
+            name.to_string(),
+            Item::new(Box::new(RecursiveFitter::new(kernel))),
+        );
+    }
+
+    pub fn contains_item(&self, name: &str) -> bool {
+        self.storage.contains_key(name)
+    }
+
+    pub fn item_score(&self, name: &str, t: f64) -> (f64, f64) {
+        let id = self.storage.get_id(name);
+        let (ms, vs) = self.storage.item(id).fitter.predict(&[t]);
+
+        (ms[0], vs[0])
+    }
+
+    pub fn observe(
+        &mut self,
+        winners: &[&str],
+        losers: &[&str],
+        diff: f64,
+        t: f64,
+        var: Option<f64>,
+    ) {
+        if t < self.last_t {
+            panic!("observations must be added in chronological order");
+        }
+
+        let var = var.unwrap_or_else(|| self.var);
+
+        let mut elems = self.process_items(winners, 1.0);
+        elems.extend(self.process_items(losers, -1.0));
+
+        let obs = GaussianObservation::new(&mut self.storage, &elems, diff, t, var);
+
+        self.observations.push(obs);
+
+        self.last_t = t;
+    }
+
+    pub fn fit(&mut self) -> bool {
+        unimplemented!();
+    }
+
+    pub fn probabilities(
+        &mut self,
+        team_1: &[&str],
+        team_2: &[&str],
+        t: f64,
+        margin: Option<f64>,
+    ) -> (f64, f64, f64) {
+        unimplemented!();
+    }
+
+    fn process_items(&self, items: &[&str], sign: f64) -> Vec<(usize, f64)> {
+        items
+            .iter()
+            .map(|key| (self.storage.get_id(&key), sign))
+            .collect()
+    }
+}
diff --git a/src/observation.rs b/src/observation.rs
index f4bc284..5582fb4 100644
--- a/src/observation.rs
+++ b/src/observation.rs
@@ -1,7 +1,9 @@
 use crate::storage::Storage;
 
+mod gaussian;
 mod ordinal;
 
+pub use gaussian::*;
 pub use ordinal::*;
 
 pub trait Observation {
diff --git a/src/observation/gaussian.rs b/src/observation/gaussian.rs
new file mode 100644
index 0000000..38f470b
--- /dev/null
+++ b/src/observation/gaussian.rs
@@ -0,0 +1,36 @@
+use crate::storage::Storage;
+
+use super::Observation;
+
+pub struct GaussianObservation {
+    m: usize,
+    items: Vec<usize>,
+    coeffs: Vec<f64>,
+    indices: Vec<usize>,
+    ns_cav: Vec<f64>,
+    xs_cav: Vec<f64>,
+    t: f64,
+    logpart: f64,
+    exp_ll: usize,
+    margin: f64,
+}
+
+impl GaussianObservation {
+    pub fn new(storage: &mut Storage, elems: &[(usize, f64)], diff: f64, t: f64, var: f64) -> Self {
+        unimplemented!();
+    }
+}
+
+impl Observation for GaussianObservation {
+    fn match_moments(&self, mean_cav: f64, cov_cav: f64) -> (f64, f64, f64) {
+        unimplemented!();
+    }
+
+    fn ep_update(&mut self, lr: f64, storage: &mut Storage) -> f64 {
+        unimplemented!();
+    }
+
+    fn kl_update(&mut self, lr: f64, storage: &mut Storage) -> f64 {
+        unimplemented!();
+    }
+}