fix(reports): split startup health latency

src/evaluator.mjs

@@ -92,7 +92,7 @@ export function evaluateRecord(record, scenario, options = {}) {
   });
   const finalGatewayState = record.finalMetrics?.service?.gatewayState ?? null;
   const healthFailures = countHealthFailures(record);
-  const healthP95Ms = collectHealthP95(record);
+  const healthP95 = collectHealthP95Evidence(record);
   const soakEvidence = collectSoakEvidence(allResults);
   const mcpBridgeEvidence = collectMcpBridgeEvidence(allResults);
   const browserAutomationEvidence = collectBrowserAutomationEvidence(allResults);
@@ -191,13 +191,15 @@ export function evaluateRecord(record, scenario, options = {}) {
     });
   }
 
-  if (typeof thresholds.healthP95Ms === "number" && healthP95Ms !== null && healthP95Ms > thresholds.healthP95Ms) {
+  if (typeof thresholds.healthP95Ms === "number" && healthP95.max !== null && healthP95.max > thresholds.healthP95Ms) {
+    const healthScope = healthP95.maxPhaseKind === "startup" ? "startup" : "post-ready";
+    const phaseSuffix = healthP95.maxPhaseId ? ` during ${healthP95.maxPhaseId}` : "";
     violations.push({
       kind: "health",
       metric: "healthP95Ms",
       expected: `<= ${thresholds.healthP95Ms}`,
-      actual: healthP95Ms,
-      message: `gateway health p95 ${healthP95Ms}ms exceeded threshold ${thresholds.healthP95Ms}ms`
+      actual: healthP95.max,
+      message: `gateway ${healthScope} health p95 ${healthP95.max}ms exceeded threshold ${thresholds.healthP95Ms}ms${phaseSuffix}`
     });
   }
 
@@ -730,7 +732,11 @@ export function evaluateRecord(record, scenario, options = {}) {
     missingDependencyErrors,
     finalGatewayState,
     healthFailures,
-    healthP95Ms,
+    healthP95Ms: healthP95.max,
+    startupHealthP95Ms: healthP95.startup,
+    postReadyHealthP95Ms: healthP95.postReady,
+    healthP95PhaseId: healthP95.maxPhaseId,
+    healthP95PhaseKind: healthP95.maxPhaseKind,
     soakEvidence,
     mcpBridgeEvidence,
     mcpInitializeMs: mcpBridgeEvidence.initializeMs,
@@ -1851,24 +1857,63 @@ function countGatewayRestarts(record) {
   return commandRestarts + countLogMetric(record, "gatewayRestartMentions");
 }
 
-function collectHealthP95(record) {
-  const p95Values = [];
+const STARTUP_HEALTH_PHASE_IDS = new Set([
+  "baseline",
+  "cold-start",
+  "gateway",
+  "gateway-start",
+  "post-upgrade",
+  "provision",
+  "restart",
+  "source-runtime",
+  "start",
+  "upgrade",
+  "warm-restart"
+]);
+
+function collectHealthP95Evidence(record) {
+  const entries = [];
   for (const phase of record.phases ?? []) {
     const p95 = phase.metrics?.healthSummary?.p95Ms;
     if (typeof p95 === "number") {
-      p95Values.push(p95);
+      entries.push({
+        phaseId: phase.id ?? null,
+        phaseKind: classifyHealthPhase(phase.id),
+        p95
+      });
     }
   }
 
   const finalP95 = record.finalMetrics?.healthSummary?.p95Ms;
   if (typeof finalP95 === "number") {
-    p95Values.push(finalP95);
+    entries.push({
+      phaseId: "final",
+      phaseKind: "post-ready",
+      p95: finalP95
+    });
   }
 
-  if (p95Values.length === 0) {
-    return null;
+  if (entries.length === 0) {
+    return {
+      max: null,
+      startup: null,
+      postReady: null,
+      maxPhaseId: null,
+      maxPhaseKind: null
+    };
   }
-  return Math.max(...p95Values);
+  const maxEntry = entries.reduce((max, entry) => (entry.p95 > max.p95 ? entry : max), entries[0]);
+  return {
+    max: maxEntry.p95,
+    startup: maxNullable(...entries.filter((entry) => entry.phaseKind === "startup").map((entry) => entry.p95)),
+    postReady: maxNullable(...entries.filter((entry) => entry.phaseKind === "post-ready").map((entry) => entry.p95)),
+    maxPhaseId: maxEntry.phaseId,
+    maxPhaseKind: maxEntry.phaseKind
+  };
+}
+
+function classifyHealthPhase(phaseId) {
+  return STARTUP_HEALTH_PHASE_IDS.has(phaseId) ? "startup" : "post-ready";
 }
 
 function collectSoakEvidence(results) {

src/performance/stats.mjs

@@ -19,7 +19,9 @@ export const PERFORMANCE_METRICS = [
   { id: "agentCleanupMaxMs", title: "Agent Cleanup Max", unit: "ms", regressionKey: "agentLatencyRegressionPercent" },
   { id: "coldPreProviderMs", title: "Cold Pre-Provider", unit: "ms", regressionKey: "agentLatencyRegressionPercent" },
   { id: "warmPreProviderMs", title: "Warm Pre-Provider", unit: "ms", regressionKey: "agentLatencyRegressionPercent" },
-  { id: "healthP95Ms", title: "Health p95", unit: "ms", regressionKey: "startupRegressionPercent" },
+  { id: "healthP95Ms", title: "Phase Health p95", unit: "ms", regressionKey: "startupRegressionPercent" },
+  { id: "startupHealthP95Ms", title: "Startup Health p95", unit: "ms", regressionKey: "startupRegressionPercent" },
+  { id: "postReadyHealthP95Ms", title: "Post-Ready Health p95", unit: "ms", regressionKey: "eventLoopRegressionPercent" },
   { id: "runtimeDepsStagingMs", title: "Runtime Deps Staging", unit: "ms", regressionKey: "startupRegressionPercent" }
 ];
 

src/reporting/compare.mjs

@@ -405,6 +405,8 @@ function metricDeltas(baseline, current) {
     "timeToListeningMs",
     "timeToHealthReadyMs",
     "healthP95Ms",
+    "startupHealthP95Ms",
+    "postReadyHealthP95Ms",
     "healthFailures",
     "readinessFailures",
     "missingDependencyErrors",

src/reporting/report.mjs

@@ -115,7 +115,12 @@ export function renderMarkdownReport(report) {
       lines.push(`- Missing dependency errors: ${record.measurements.missingDependencyErrors ?? "unknown"}`);
       lines.push(`- Final gateway state: ${record.measurements.finalGatewayState ?? "unknown"}`);
       lines.push(`- Health failures: ${record.measurements.healthFailures ?? "unknown"}`);
-      lines.push(`- Health p95: ${record.measurements.healthP95Ms ?? "unknown"} ms`);
+      lines.push(`- Phase health p95: ${record.measurements.healthP95Ms ?? "unknown"} ms`);
+      lines.push(`- Startup health p95: ${record.measurements.startupHealthP95Ms ?? "unknown"} ms`);
+      lines.push(`- Post-ready health p95: ${record.measurements.postReadyHealthP95Ms ?? "unknown"} ms`);
+      if (record.measurements.healthP95PhaseId) {
+        lines.push(`- Slowest health phase: ${record.measurements.healthP95PhaseId} (${record.measurements.healthP95PhaseKind ?? "unknown"})`);
+      }
       if (record.measurements.soakEvidence?.available) {
         lines.push(`- Soak trend: duration ${record.measurements.soakDurationMs ?? "unknown"} ms; iterations ${record.measurements.soakIterations ?? "unknown"}; command p95 ${record.measurements.soakCommandP95Ms ?? "unknown"} ms; health p95 ${record.measurements.soakHealthP95Ms ?? "unknown"} ms; RSS growth ${record.measurements.rssGrowthMb ?? "unknown"} MB; gateway RSS growth ${record.measurements.gatewayRssGrowthMb ?? "unknown"} MB`);
       }

src/selfcheck.mjs

@@ -338,6 +338,7 @@ export async function runSelfCheck(flags = {}) {
     checks.push(markdownFailureCardsCheck());
     checks.push(reportRecommendedNextScenarioCheck());
     checks.push(readinessClassificationCheck());
+    checks.push(healthP95SplitEvaluationCheck());
     checks.push(await resourceRoleAttributionCheck(tmp));
     checks.push(await resourceRootCommandRoleBoundaryCheck());
     checks.push(await resourceRolePollutionCheck());
@@ -3648,6 +3649,69 @@ function readinessClassificationCheck() {
   }
 }
 
+function healthP95SplitEvaluationCheck() {
+  try {
+    const record = {
+      scenario: "gateway-performance",
+      title: "Gateway Performance",
+      status: "PASS",
+      phases: [
+        {
+          id: "cold-start",
+          title: "Cold start",
+          results: [],
+          metrics: {
+            logs: zeroLogMetrics(),
+            health: { ok: true },
+            healthSummary: { count: 3, okCount: 3, failureCount: 0, p95Ms: 30, maxMs: 30 }
+          }
+        },
+        {
+          id: "api-latency",
+          title: "API latency",
+          results: [],
+          metrics: {
+            logs: zeroLogMetrics(),
+            health: { ok: true },
+            healthSummary: { count: 3, okCount: 3, failureCount: 0, p95Ms: 220, maxMs: 220 }
+          }
+        }
+      ],
+      finalMetrics: {
+        service: { gatewayState: "running" },
+        logs: zeroLogMetrics(),
+        healthSummary: { count: 3, okCount: 3, failureCount: 0, p95Ms: 45, maxMs: 45 }
+      }
+    };
+    evaluateRecord(record, {
+      id: "gateway-performance",
+      thresholds: { healthP95Ms: 100 }
+    }, { surface: { thresholds: {} }, targetPlan: { kind: "local-build" } });
+
+    assertEqual(record.measurements.healthP95Ms, 220, "phase health p95");
+    assertEqual(record.measurements.startupHealthP95Ms, 30, "startup health p95");
+    assertEqual(record.measurements.postReadyHealthP95Ms, 220, "post-ready health p95");
+    assertEqual(record.measurements.healthP95PhaseId, "api-latency", "health p95 phase id");
+    assertEqual(record.measurements.healthP95PhaseKind, "post-ready", "health p95 phase kind");
+    assertEqual(record.violations.some((violation) => violation.message.includes("post-ready health p95") && violation.message.includes("api-latency")), true, "health violation names post-ready phase");
+
+    return {
+      id: "health-p95-split",
+      status: "PASS",
+      command: "evaluate health p95 startup/post-ready split",
+      durationMs: 0
+    };
+  } catch (error) {
+    return {
+      id: "health-p95-split",
+      status: "FAIL",
+      command: "evaluate health p95 startup/post-ready split",
+      durationMs: 0,
+      message: error.message
+    };
+  }
+}
+
 async function resourceRoleAttributionCheck(tmp) {
   const command = "node -e 'setTimeout(() => {}, 650)'";
   const artifactPath = join(tmp, "resource-role-attribution.jsonl");