diff --git a/src/truth/ecan.rs b/src/truth/ecan.rs
new file mode 100644
index 0000000..144b437
--- /dev/null
+++ b/src/truth/ecan.rs
@@ -0,0 +1,274 @@
+//! Economic Attention Network (ECAN) for memory importance management.
+//!
+//! ECAN assigns Short-Term Importance (STI) and Long-Term Importance (LTI)
+//! to memories, enabling natural prioritization of verified, frequently-confirmed
+//! knowledge while deprioritizing stale or unreliable memories.
+//!
+//! Based on the OpenCog ECAN model adapted for the Bushidai Truth Engine.
+
+/// Clamp a value to the range [0.0, 1.0].
+#[inline]
+fn clamp01(v: f32) -> f32 {
+    v.clamp(0.0, 1.0)
+}
+
+/// ECAN parameters controlling attention dynamics.
+#[derive(Debug, Clone)]
+pub struct EcanParams {
+    /// STI decay rate per cycle (0.0–1.0). Higher = slower decay.
+    pub decay_rate: f32,
+    /// LTI update rate — how fast LTI accumulates from STI.
+    pub beta: f32,
+    /// Fraction of STI that spreads to LTI each cycle.
+    pub spread_factor: f32,
+    /// STI threshold below which LTI update is penalized.
+    pub threshold: f32,
+}
+
+impl EcanParams {
+    /// Create EcanParams with the given decay rate and spread factor.
+    /// Other parameters use sensible defaults.
+    pub fn new(decay_rate: f32, spread_factor: f32) -> Self {
+        Self {
+            decay_rate: clamp01(decay_rate),
+            beta: 0.01,
+            spread_factor: clamp01(spread_factor),
+            threshold: 0.1,
+        }
+    }
+
+    /// Apply STI decay. Each cycle, STI moves toward zero.
+    ///
+    /// ```text
+    /// new_sti = sti * decay_rate
+    /// ```
+    pub fn decay_sti(&self, sti: f32) -> f32 {
+        clamp01(sti * self.decay_rate)
+    }
+
+    /// Update LTI based on current STI.
+    ///
+    /// LTI accumulates slowly when STI is high (memory is actively relevant).
+    /// When STI falls below threshold, LTI update is penalized.
+    ///
+    /// ```text
+    /// new_lti = lti * (1 - beta) + spread_factor * sti
+    /// if sti < threshold: new_lti *= 0.5
+    /// ```
+    pub fn update_lti(&self, lti: f32, sti: f32) -> f32 {
+        let mut new_lti = lti * (1.0 - self.beta) + self.spread_factor * sti;
+        if sti < self.threshold {
+            new_lti *= 0.5;
+        }
+        clamp01(new_lti)
+    }
+
+    /// Boost STI by a spread amount (e.g., when a memory is confirmed).
+    pub fn spread(&self, sti: f32, amount: f32) -> f32 {
+        clamp01(sti + amount)
+    }
+
+    /// Run a full ECAN cycle: decay STI, then update LTI.
+    ///
+    /// The truth_value is used to weight the STI contribution:
+    /// higher truth values contribute more to LTI accumulation.
+    pub fn cycle(&self, sti: f32, lti: f32, truth_value: f32) -> (f32, f32) {
+        let new_sti = self.decay_sti(sti);
+        // Weight STI contribution by truth value for LTI update
+        let weighted_sti = new_sti * clamp01(truth_value);
+        let new_lti = self.update_lti(lti, weighted_sti);
+        (new_sti, new_lti)
+    }
+
+    /// Compute initial STI and LTI for a newly scored memory.
+    ///
+    /// STI starts proportional to truth_value (high truth = high initial attention).
+    /// LTI starts proportional to confidence (high confidence = more reliable).
+    pub fn initialize(&self, truth_value: f32, confidence: f32) -> (f32, f32) {
+        let sti = clamp01(truth_value);
+        let lti = clamp01(confidence * 0.5); // Start LTI conservatively
+        (sti, lti)
+    }
+}
+
+impl Default for EcanParams {
+    fn default() -> Self {
+        Self {
+            decay_rate: 0.95,
+            beta: 0.01,
+            spread_factor: 0.05,
+            threshold: 0.1,
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn default_params() -> EcanParams {
+        EcanParams::default()
+    }
+
+    #[test]
+    fn ecan_decay_reduces_sti() {
+        let params = default_params();
+        let sti = 0.8;
+        let decayed = params.decay_sti(sti);
+        assert!(decayed < sti, "STI should decrease after decay");
+        assert!(decayed > 0.0, "STI should remain positive");
+    }
+
+    #[test]
+    fn ecan_decay_converges_to_zero() {
+        let params = default_params();
+        let mut sti = 1.0;
+        for _ in 0..200 {
+            sti = params.decay_sti(sti);
+        }
+        assert!(
+            sti < 0.001,
+            "STI should converge near zero after many cycles, got {}",
+            sti
+        );
+    }
+
+    #[test]
+    fn ecan_lti_grows_with_high_sti() {
+        let params = default_params();
+        let lti = 0.1;
+        let high_sti = 0.9;
+        let new_lti = params.update_lti(lti, high_sti);
+        assert!(
+            new_lti > lti,
+            "LTI should grow when STI is high: {} > {}",
+            new_lti,
+            lti
+        );
+    }
+
+    #[test]
+    fn ecan_lti_penalized_below_threshold() {
+        let params = default_params();
+        let lti = 0.5;
+        let low_sti = 0.01; // Below threshold of 0.1
+        let new_lti = params.update_lti(lti, low_sti);
+        // With penalty, LTI should be roughly halved
+        assert!(
+            new_lti < lti * 0.6,
+            "LTI should be penalized below threshold: {} < {}",
+            new_lti,
+            lti * 0.6
+        );
+    }
+
+    #[test]
+    fn ecan_lti_not_penalized_above_threshold() {
+        let params = default_params();
+        let lti = 0.5;
+        let above_threshold_sti = 0.5;
+        let new_lti = params.update_lti(lti, above_threshold_sti);
+        // LTI should stay close to original or grow slightly
+        assert!(
+            new_lti >= lti * 0.9,
+            "LTI should not be heavily penalized above threshold: {} >= {}",
+            new_lti,
+            lti * 0.9
+        );
+    }
+
+    #[test]
+    fn ecan_spread_increases_sti() {
+        let params = default_params();
+        let sti = 0.5;
+        let boosted = params.spread(sti, 0.2);
+        assert_eq!(boosted, 0.7, "Spread should add to STI");
+    }
+
+    #[test]
+    fn ecan_spread_clamps_to_one() {
+        let params = default_params();
+        let sti = 0.9;
+        let boosted = params.spread(sti, 0.5);
+        assert_eq!(boosted, 1.0, "Spread should clamp at 1.0");
+    }
+
+    #[test]
+    fn ecan_full_cycle() {
+        let params = default_params();
+        let (new_sti, new_lti) = params.cycle(0.8, 0.3, 0.9);
+        // STI should decay
+        assert!(new_sti < 0.8, "STI should decay in cycle");
+        assert!(new_sti > 0.0, "STI should remain positive");
+        // LTI should adjust based on weighted STI
+        assert!(new_lti > 0.0, "LTI should remain positive");
+        // Both bounded
+        assert!(new_sti <= 1.0 && new_lti <= 1.0);
+    }
+
+    #[test]
+    fn ecan_cycle_low_truth_reduces_lti_contribution() {
+        let params = default_params();
+        let (_, lti_high_truth) = params.cycle(0.8, 0.3, 0.9);
+        let (_, lti_low_truth) = params.cycle(0.8, 0.3, 0.1);
+        // Higher truth value should contribute more to LTI
+        assert!(
+            lti_high_truth >= lti_low_truth,
+            "High truth should contribute more to LTI: {} >= {}",
+            lti_high_truth,
+            lti_low_truth
+        );
+    }
+
+    #[test]
+    fn ecan_initialize_from_truth_value() {
+        let params = default_params();
+        let (sti, lti) = params.initialize(0.9, 0.8);
+        assert_eq!(sti, 0.9, "Initial STI should equal truth_value");
+        assert!(
+            (lti - 0.4).abs() < 0.001,
+            "Initial LTI should be confidence * 0.5 = 0.4, got {}",
+            lti
+        );
+    }
+
+    #[test]
+    fn ecan_initialize_zero() {
+        let params = default_params();
+        let (sti, lti) = params.initialize(0.0, 0.0);
+        assert_eq!(sti, 0.0);
+        assert_eq!(lti, 0.0);
+    }
+
+    #[test]
+    fn ecan_values_bounded() {
+        let params = default_params();
+        // Test with extreme values
+        let decayed = params.decay_sti(1.5);
+        assert!(decayed <= 1.0, "Decay should clamp to 1.0");
+
+        let lti = params.update_lti(2.0, 2.0);
+        assert!(lti <= 1.0, "LTI update should clamp to 1.0");
+
+        let spread = params.spread(0.8, 0.5);
+        assert!(spread <= 1.0, "Spread should clamp to 1.0");
+
+        let (sti, lti) = params.cycle(1.5, 1.5, 1.5);
+        assert!(sti <= 1.0 && lti <= 1.0, "Cycle should clamp outputs");
+
+        let (sti, lti) = params.initialize(2.0, 2.0);
+        assert!(
+            sti <= 1.0 && lti <= 1.0,
+            "Initialize should clamp outputs"
+        );
+    }
+
+    #[test]
+    fn ecan_from_new() {
+        let params = EcanParams::new(0.90, 0.10);
+        assert_eq!(params.decay_rate, 0.90);
+        assert_eq!(params.spread_factor, 0.10);
+        assert_eq!(params.beta, 0.01);
+        assert_eq!(params.threshold, 0.1);
+    }
+}
diff --git a/src/truth/mod.rs b/src/truth/mod.rs
index 76ce9fe..1204c63 100644
--- a/src/truth/mod.rs
+++ b/src/truth/mod.rs
@@ -1,3 +1,15 @@
 //! Truth scoring engine — PLN deduction, ECAN attention, and memory scoring.
+//!
+//! This module implements neuro-symbolic reasoning for evaluating the
+//! truthfulness of stored memories. Based on the Bushidai Truth Simulator
+//! by Thijs Smits (TS87).
+//!
+//! ## Components
+//!
+//! - **PLN** (Probabilistic Logic Networks): Deduction rules for computing
+//!   truth values from evidence chains.
+//! - **ECAN** (Economic Attention Network): Manages short-term and long-term
+//!   importance of memories, enabling natural prioritization of verified knowledge.
 
+pub mod ecan;
 pub mod pln;