diff --git a/src/lib.rs b/src/lib.rs
index f4b3909..52be22a 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -7,6 +7,7 @@ pub mod embedding;
 pub mod migrations;
 pub mod tools;
 pub mod transport;
+pub mod truth;
 pub mod ttl;
 
 use anyhow::Result;
diff --git a/src/truth/mod.rs b/src/truth/mod.rs
new file mode 100644
index 0000000..76ce9fe
--- /dev/null
+++ b/src/truth/mod.rs
@@ -0,0 +1,3 @@
+//! Truth scoring engine — PLN deduction, ECAN attention, and memory scoring.
+
+pub mod pln;
diff --git a/src/truth/pln.rs b/src/truth/pln.rs
new file mode 100644
index 0000000..dd0812c
--- /dev/null
+++ b/src/truth/pln.rs
@@ -0,0 +1,319 @@
+//! Probabilistic Logic Networks (PLN) deduction engine.
+//!
+//! Implements core PLN inference rules for truth-value propagation:
+//! deduction, revision, negation, conjunction, and evidence-based scoring.
+
+/// A PLN truth value combining strength (probability estimate) and confidence
+/// (weight of evidence behind that estimate). Both fields are clamped to \[0.0, 1.0\].
+#[derive(Debug, Clone, Copy, PartialEq)]
+pub struct TruthValue {
+    /// Probability estimate, 0.0–1.0.
+    pub strength: f32,
+    /// Weight of evidence, 0.0–1.0.
+    pub confidence: f32,
+}
+
+impl TruthValue {
+    /// Create a new `TruthValue`, clamping both fields to \[0.0, 1.0\].
+    pub fn new(strength: f32, confidence: f32) -> Self {
+        Self {
+            strength: strength.clamp(0.0, 1.0),
+            confidence: confidence.clamp(0.0, 1.0),
+        }
+    }
+}
+
+/// Clamp a `TruthValue` so both fields lie within \[0.0, 1.0\].
+#[inline]
+fn clamped(tv: TruthValue) -> TruthValue {
+    TruthValue::new(tv.strength, tv.confidence)
+}
+
+// ---------------------------------------------------------------------------
+// PLN Rules
+// ---------------------------------------------------------------------------
+
+/// **Deduction rule** — chain two implications A→B and B→C.
+///
+/// ```text
+/// strength  = s1 * s2
+/// confidence = c1 * c2 * (s2 + (1 - s2) * (1 - c1))
+/// ```
+pub fn deduction(a_to_b: TruthValue, b_to_c: TruthValue) -> TruthValue {
+    let s1 = a_to_b.strength;
+    let c1 = a_to_b.confidence;
+    let s2 = b_to_c.strength;
+    let c2 = b_to_c.confidence;
+
+    let strength = s1 * s2;
+    let confidence = c1 * c2 * (s2 + (1.0 - s2) * (1.0 - c1));
+
+    clamped(TruthValue { strength, confidence })
+}
+
+/// **Revision rule** — merge two independent estimates into one.
+///
+/// ```text
+/// strength  = (s1*c1 + s2*c2) / (c1 + c2)
+/// confidence = (c1 + c2) / (c1 + c2 + 1)
+/// ```
+///
+/// If both confidences are zero the input `tv1` is returned unchanged
+/// (no evidence to merge).
+pub fn revision(tv1: TruthValue, tv2: TruthValue) -> TruthValue {
+    let s1 = tv1.strength;
+    let c1 = tv1.confidence;
+    let s2 = tv2.strength;
+    let c2 = tv2.confidence;
+
+    let denom = c1 + c2;
+    if denom == 0.0 {
+        return tv1;
+    }
+
+    let strength = (s1 * c1 + s2 * c2) / denom;
+    let confidence = denom / (denom + 1.0);
+
+    clamped(TruthValue { strength, confidence })
+}
+
+/// **Negation rule** — logical NOT.
+///
+/// ```text
+/// strength  = 1.0 - s
+/// confidence = c          (unchanged)
+/// ```
+pub fn negation(tv: TruthValue) -> TruthValue {
+    clamped(TruthValue {
+        strength: 1.0 - tv.strength,
+        confidence: tv.confidence,
+    })
+}
+
+/// **Conjunction rule** — logical AND (independent assumptions).
+///
+/// ```text
+/// strength  = s1 * s2
+/// confidence = min(c1, c2)
+/// ```
+pub fn conjunction(tv1: TruthValue, tv2: TruthValue) -> TruthValue {
+    clamped(TruthValue {
+        strength: tv1.strength * tv2.strength,
+        confidence: tv1.confidence.min(tv2.confidence),
+    })
+}
+
+/// **Evidence-based scoring** — refine a base truth value with confirming and
+/// contradicting evidence.
+///
+/// 1. Iteratively apply [`revision`] across all `confirmations` to boost the
+///    base estimate.
+/// 2. For each contradiction, reduce confidence proportionally to the
+///    contradiction's strength.
+/// 3. Clamp the final result to \[0.0, 1.0\].
+pub fn score_with_evidence(
+    base: TruthValue,
+    confirmations: &[TruthValue],
+    contradictions: &[TruthValue],
+) -> TruthValue {
+    // Phase 1: merge confirmations via revision
+    let mut current = base;
+    for &conf in confirmations {
+        current = revision(current, conf);
+    }
+
+    // Phase 2: reduce confidence proportional to contradiction strength
+    for &contra in contradictions {
+        // Each contradiction chips away at confidence in proportion to its
+        // own strength (strong contradictions hurt more).
+        current.confidence *= 1.0 - contra.strength * contra.confidence;
+    }
+
+    clamped(current)
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// Helper: approximate f32 equality.
+    fn approx_eq(a: f32, b: f32) -> bool {
+        (a - b).abs() < 1e-5
+    }
+
+    // -- deduction ---------------------------------------------------------
+
+    #[test]
+    fn pln_deduction_basic() {
+        let ab = TruthValue::new(0.8, 0.9);
+        let bc = TruthValue::new(0.7, 0.8);
+        let result = deduction(ab, bc);
+
+        // strength = 0.8 * 0.7 = 0.56
+        assert!(approx_eq(result.strength, 0.56));
+
+        // confidence = 0.9 * 0.8 * (0.7 + 0.3 * 0.1) = 0.72 * 0.73 = 0.5256
+        let expected_c = 0.9 * 0.8 * (0.7 + (1.0 - 0.7) * (1.0 - 0.9));
+        assert!(approx_eq(result.confidence, expected_c));
+    }
+
+    #[test]
+    fn pln_deduction_boundary() {
+        // (0, 0) -> zero everything
+        let r1 = deduction(TruthValue::new(0.0, 0.0), TruthValue::new(0.0, 0.0));
+        assert!(approx_eq(r1.strength, 0.0));
+        assert!(approx_eq(r1.confidence, 0.0));
+
+        // (1, 1) -> perfect chain
+        let r2 = deduction(TruthValue::new(1.0, 1.0), TruthValue::new(1.0, 1.0));
+        assert!(approx_eq(r2.strength, 1.0));
+        // confidence = 1*1*(1 + 0*0) = 1.0
+        assert!(approx_eq(r2.confidence, 1.0));
+
+        // (0.5, 0.5)
+        let half = TruthValue::new(0.5, 0.5);
+        let r3 = deduction(half, half);
+        assert!(approx_eq(r3.strength, 0.25));
+        let expected_c = 0.5 * 0.5 * (0.5 + 0.5 * 0.5);
+        assert!(approx_eq(r3.confidence, expected_c));
+    }
+
+    // -- revision ----------------------------------------------------------
+
+    #[test]
+    fn pln_revision_merge() {
+        let tv1 = TruthValue::new(0.6, 0.5);
+        let tv2 = TruthValue::new(0.8, 0.7);
+        let result = revision(tv1, tv2);
+
+        let expected_s = (0.6 * 0.5 + 0.8 * 0.7) / (0.5 + 0.7);
+        let expected_c = (0.5 + 0.7) / (0.5 + 0.7 + 1.0);
+        assert!(approx_eq(result.strength, expected_s));
+        assert!(approx_eq(result.confidence, expected_c));
+    }
+
+    #[test]
+    fn pln_revision_symmetry() {
+        let a = TruthValue::new(0.3, 0.4);
+        let b = TruthValue::new(0.9, 0.6);
+
+        let ab = revision(a, b);
+        let ba = revision(b, a);
+
+        assert!(approx_eq(ab.strength, ba.strength));
+        assert!(approx_eq(ab.confidence, ba.confidence));
+    }
+
+    // -- negation ----------------------------------------------------------
+
+    #[test]
+    fn pln_negation() {
+        let tv = TruthValue::new(0.7, 0.8);
+        let result = negation(tv);
+
+        assert!(approx_eq(result.strength, 0.3));
+        assert!(approx_eq(result.confidence, 0.8));
+    }
+
+    // -- conjunction -------------------------------------------------------
+
+    #[test]
+    fn pln_conjunction() {
+        let tv1 = TruthValue::new(0.8, 0.9);
+        let tv2 = TruthValue::new(0.6, 0.7);
+        let result = conjunction(tv1, tv2);
+
+        assert!(approx_eq(result.strength, 0.48));
+        assert!(approx_eq(result.confidence, 0.7));
+    }
+
+    // -- score_with_evidence -----------------------------------------------
+
+    #[test]
+    fn pln_score_with_confirmations() {
+        let base = TruthValue::new(0.5, 0.3);
+        let confirmations = [
+            TruthValue::new(0.7, 0.5),
+            TruthValue::new(0.8, 0.6),
+        ];
+
+        let result = score_with_evidence(base, &confirmations, &[]);
+
+        // Confirmations should boost both strength and confidence above base
+        assert!(result.strength > base.strength);
+        assert!(result.confidence > base.confidence);
+        assert!(result.strength <= 1.0);
+        assert!(result.confidence <= 1.0);
+    }
+
+    #[test]
+    fn pln_score_with_contradictions() {
+        let base = TruthValue::new(0.8, 0.9);
+        let contradictions = [
+            TruthValue::new(0.7, 0.8),
+        ];
+
+        let result = score_with_evidence(base, &[], &contradictions);
+
+        // Contradictions should reduce confidence
+        assert!(result.confidence < base.confidence);
+        // Strength unchanged when no confirmations
+        assert!(approx_eq(result.strength, base.strength));
+    }
+
+    // -- bounds ------------------------------------------------------------
+
+    #[test]
+    fn pln_values_bounded() {
+        // Extreme inputs that might push results out of range
+        let big = TruthValue::new(2.0, 5.0);
+        assert!(big.strength <= 1.0 && big.strength >= 0.0);
+        assert!(big.confidence <= 1.0 && big.confidence >= 0.0);
+
+        let neg = TruthValue::new(-1.0, -3.0);
+        assert!(neg.strength >= 0.0);
+        assert!(neg.confidence >= 0.0);
+
+        // Deduction of clamped extremes stays in bounds
+        let d = deduction(big, big);
+        assert!(d.strength >= 0.0 && d.strength <= 1.0);
+        assert!(d.confidence >= 0.0 && d.confidence <= 1.0);
+
+        // Negation of boundary stays in bounds
+        let n = negation(TruthValue::new(0.0, 1.0));
+        assert!(n.strength >= 0.0 && n.strength <= 1.0);
+        assert!(n.confidence >= 0.0 && n.confidence <= 1.0);
+
+        // score_with_evidence stays bounded
+        let scored = score_with_evidence(
+            TruthValue::new(1.0, 1.0),
+            &[TruthValue::new(1.0, 1.0)],
+            &[TruthValue::new(1.0, 1.0)],
+        );
+        assert!(scored.strength >= 0.0 && scored.strength <= 1.0);
+        assert!(scored.confidence >= 0.0 && scored.confidence <= 1.0);
+    }
+
+    // -- zero-confidence edge case -----------------------------------------
+
+    #[test]
+    fn pln_revision_zero_confidence() {
+        let tv1 = TruthValue::new(0.5, 0.0);
+        let tv2 = TruthValue::new(0.8, 0.0);
+
+        // Should not panic; returns tv1 when both confidences are zero
+        let result = revision(tv1, tv2);
+        assert!(approx_eq(result.strength, tv1.strength));
+        assert!(approx_eq(result.confidence, tv1.confidence));
+
+        // One zero, one nonzero — should still work
+        let tv3 = TruthValue::new(0.9, 0.6);
+        let result2 = revision(tv1, tv3);
+        assert!(result2.strength >= 0.0 && result2.strength <= 1.0);
+        assert!(result2.confidence >= 0.0 && result2.confidence <= 1.0);
+    }
+}