perf: expose and reduce message publish latency

loadtest/im-pressure/merge_test.go

@@ -80,6 +80,37 @@ func TestMergedAckP95FallsBackWhenRawSamplesAreIncomplete(t *testing.T) {
 	}
 }
 
+func TestRunMergeUsesSameClockDeliveryLatencyForGate(t *testing.T) {
+	dir := t.TempDir()
+	local := HistogramSnapshot{Count: 10, MaxMS: 200, Buckets: map[string]int64{"200ms": 10}}
+	cross := HistogramSnapshot{Count: 10, MaxMS: 5000, Buckets: map[string]int64{"5s": 10}}
+	a := AgentSummary{RunID: "run1", Scenario: "p2p", Machine: "a", ExpectedConnections: 1,
+		ReadyConnections: 1, PublishAttempted: 1, PublishAccepted: 1, ExpectedPeer: 1, ReceivedPeer: 1,
+		DeliveryLatencyLocal: local, DeliveryLatencyCross: cross, CrossClockSkewSamples: 3}
+	b := a
+	b.Machine = "b"
+	aPath, bPath, reportPath := filepath.Join(dir, "a.json"), filepath.Join(dir, "b.json"), filepath.Join(dir, "report.json")
+	writeSummaryForTest(t, aPath, a)
+	writeSummaryForTest(t, bPath, b)
+	if err := runMerge(aPath, bPath, reportPath); err != nil {
+		t.Fatal(err)
+	}
+	var report CombinedReport
+	raw, err := os.ReadFile(reportPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err = json.Unmarshal(raw, &report); err != nil {
+		t.Fatal(err)
+	}
+	if report.DeliveryP95MS != 200 || report.DeliveryLocalP95MS != 200 || report.DeliveryCrossP95MS != 5000 {
+		t.Fatalf("unexpected delivery percentiles: %+v", report)
+	}
+	if report.CrossClockSkewSamples != 6 || !report.Passed {
+		t.Fatalf("cross-machine clock skew must not fail the same-clock latency gate: %+v", report)
+	}
+}
+
 func writeSummaryForTest(t *testing.T, path string, summary AgentSummary) {
 	t.Helper()
 	raw, err := json.Marshal(summary)