diff --git a/services/agent/governance/__init__.py b/services/agent/governance/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/services/agent/governance/audit.py b/services/agent/governance/audit.py
new file mode 100644
index 00000000..afe35876
--- /dev/null
+++ b/services/agent/governance/audit.py
@@ -0,0 +1,25 @@
+from dataclasses import dataclass, field
+from datetime import datetime, UTC
+
+
+@dataclass
+class LearningAuditRecord:
+    agent_id: str
+    event: str
+    previous_score: float
+    new_score: float
+    reason: str
+    created_at: datetime = field(
+        default_factory=lambda: datetime.now(UTC)
+    )
+
+
+class LearningAuditTrail:
+    def __init__(self):
+        self.records = []
+
+    def record(self, audit: LearningAuditRecord):
+        self.records.append(audit)
+
+    def latest(self):
+        return self.records[-1] if self.records else None
diff --git a/services/agent/governance/policy.py b/services/agent/governance/policy.py
new file mode 100644
index 00000000..32d4b3a8
--- /dev/null
+++ b/services/agent/governance/policy.py
@@ -0,0 +1,18 @@
+from dataclasses import dataclass
+
+
+@dataclass
+class LearningPolicy:
+    reward_threshold: float = 0.8
+    allow_evolution: bool = True
+
+
+class LearningPolicyEngine:
+    def __init__(self, policy: LearningPolicy | None = None):
+        self.policy = policy or LearningPolicy()
+
+    def evaluate(self, reward: float) -> bool:
+        if not self.policy.allow_evolution:
+            return False
+
+        return reward >= self.policy.reward_threshold
diff --git a/tests/agent/test_learning_governance.py b/tests/agent/test_learning_governance.py
new file mode 100644
index 00000000..9089486d
--- /dev/null
+++ b/tests/agent/test_learning_governance.py
@@ -0,0 +1,31 @@
+from services.agent.governance.audit import (
+    LearningAuditRecord,
+    LearningAuditTrail,
+)
+from services.agent.governance.policy import (
+    LearningPolicyEngine,
+)
+
+
+def test_learning_audit_record():
+    trail = LearningAuditTrail()
+
+    record = LearningAuditRecord(
+        agent_id="AEON-001",
+        event="reward_update",
+        previous_score=0.8,
+        new_score=0.95,
+        reason="accuracy_improved",
+    )
+
+    trail.record(record)
+
+    assert trail.latest().agent_id == "AEON-001"
+    assert trail.latest().new_score == 0.95
+
+
+def test_learning_policy():
+    engine = LearningPolicyEngine()
+
+    assert engine.evaluate(0.95) is True
+    assert engine.evaluate(0.5) is False