diff --git a/pkg/assertoor/testrunner.go b/pkg/assertoor/testrunner.go
index 05f54989..e5c471ff 100644
--- a/pkg/assertoor/testrunner.go
+++ b/pkg/assertoor/testrunner.go
@@ -412,6 +412,30 @@ func (c *TestRunner) RunTestCleanup(ctx context.Context, retentionTime time.Dura
 	}
 }
 
-func (c *TestRunner) cleanupTestHistory(_ time.Duration) {
-	// TODO: clean db
+// cleanupTestHistory evicts finished test runs from testRunMap once they are
+// older than retentionTime. Runs that are still pending or running are never
+// touched. Evicting a run here only drops the in-memory reference (and
+// everything it pins: its scheduler, task states, buffered loggers); the
+// persisted result stays in the database and is served from there via
+// test.WrapDBTestRun once the run is no longer in testRunMap.
+func (c *TestRunner) cleanupTestHistory(retentionTime time.Duration) {
+	cutoff := time.Now().Add(-retentionTime)
+
+	c.testRegistryMutex.Lock()
+	defer c.testRegistryMutex.Unlock()
+
+	for runID, testRef := range c.testRunMap {
+		switch testRef.Status() {
+		case types.TestStatusSuccess, types.TestStatusFailure, types.TestStatusAborted, types.TestStatusSkipped:
+		default:
+			continue
+		}
+
+		stopTime := testRef.StopTime()
+		if stopTime.IsZero() || stopTime.After(cutoff) {
+			continue
+		}
+
+		delete(c.testRunMap, runID)
+	}
 }
diff --git a/pkg/assertoor/testrunner_test.go b/pkg/assertoor/testrunner_test.go
new file mode 100644
index 00000000..78088fda
--- /dev/null
+++ b/pkg/assertoor/testrunner_test.go
@@ -0,0 +1,80 @@
+package assertoor
+
+import (
+	"testing"
+	"time"
+
+	"github.com/ethpandaops/assertoor/pkg/types"
+)
+
+// stubTest is a minimal types.Test double. Only RunID, Status, and StopTime
+// matter for cleanupTestHistory; the rest exist to satisfy the interface.
+type stubTest struct {
+	runID    uint64
+	status   types.TestStatus
+	stopTime time.Time
+}
+
+func (s *stubTest) RunID() uint64                         { return s.runID }
+func (s *stubTest) TestID() string                        { return "stub-test" }
+func (s *stubTest) Name() string                          { return "stub test" }
+func (s *stubTest) StartTime() time.Time                  { return s.stopTime.Add(-time.Minute) }
+func (s *stubTest) StopTime() time.Time                   { return s.stopTime }
+func (s *stubTest) Timeout() time.Duration                { return 0 }
+func (s *stubTest) Status() types.TestStatus              { return s.status }
+func (s *stubTest) GetTaskScheduler() types.TaskScheduler { return nil }
+func (s *stubTest) AbortTest(_ bool)                      {}
+
+func TestCleanupTestHistory(t *testing.T) {
+	const retention = time.Hour
+
+	now := time.Now()
+
+	tr := NewTestRunner(nil, 0)
+
+	tr.testRunMap[1] = &stubTest{runID: 1, status: types.TestStatusSuccess, stopTime: now.Add(-2 * retention)}
+	tr.testRunMap[2] = &stubTest{runID: 2, status: types.TestStatusFailure, stopTime: now.Add(-2 * retention)}
+	tr.testRunMap[3] = &stubTest{runID: 3, status: types.TestStatusAborted, stopTime: now.Add(-2 * retention)}
+	tr.testRunMap[4] = &stubTest{runID: 4, status: types.TestStatusSuccess, stopTime: now.Add(-retention / 2)}
+	tr.testRunMap[5] = &stubTest{runID: 5, status: types.TestStatusRunning, stopTime: time.Time{}}
+	tr.testRunMap[6] = &stubTest{runID: 6, status: types.TestStatusPending, stopTime: time.Time{}}
+
+	tr.cleanupTestHistory(retention)
+
+	wantEvicted := []uint64{1, 2, 3}
+	for _, runID := range wantEvicted {
+		if _, ok := tr.testRunMap[runID]; ok {
+			t.Errorf("run %d: expected eviction (finished more than retention ago), still present", runID)
+		}
+	}
+
+	wantKept := []uint64{4, 5, 6}
+	for _, runID := range wantKept {
+		if _, ok := tr.testRunMap[runID]; !ok {
+			t.Errorf("run %d: expected to be kept, was evicted", runID)
+		}
+	}
+
+	if got := len(tr.testRunMap); got != len(wantKept) {
+		t.Fatalf("testRunMap has %d entries after cleanup, want %d", got, len(wantKept))
+	}
+}
+
+func TestCleanupTestHistoryNeverEvictsWithoutRetention(t *testing.T) {
+	tr := NewTestRunner(nil, 0)
+
+	tr.testRunMap[1] = &stubTest{
+		runID:    1,
+		status:   types.TestStatusSuccess,
+		stopTime: time.Now().Add(-365 * 24 * time.Hour),
+	}
+
+	// A zero retention window still means "evict anything already finished",
+	// not "keep everything forever" -- RunTestCleanup is what clamps a
+	// non-positive config value to the 14 day default before calling in.
+	tr.cleanupTestHistory(0)
+
+	if _, ok := tr.testRunMap[1]; ok {
+		t.Fatalf("expected the year-old finished run to be evicted with a zero retention window")
+	}
+}