diff --git a/test/e2e/live/concurrent-gateway-ports.test.ts b/test/e2e/live/concurrent-gateway-ports.test.ts
index bc2db09f4ae..5d5fd3f5770 100644
--- a/test/e2e/live/concurrent-gateway-ports.test.ts
+++ b/test/e2e/live/concurrent-gateway-ports.test.ts
@@ -329,8 +329,12 @@ test("concurrent gateway ports: onboards two sandboxes on isolated gateways and
 
   const gatewayA = gatewayNameForPort(GATEWAY_PORT_A);
   const gatewayB = gatewayNameForPort(GATEWAY_PORT_B);
+  // OpenShell reaches this fixture from its gateway network namespace, where
+  // the runner's loopback address is not routable.
   const fake = await startFakeOpenAiCompatibleServer({
+    host: "0.0.0.0",
     port: Number(process.env.NEMOCLAW_E2E_FAKE_PORT ?? 0),
+    publicHost: "host.openshell.internal",
   });
   await artifacts.target.declare({
     id: "concurrent-gateway-ports",
diff --git a/test/e2e/live/double-onboard.test.ts b/test/e2e/live/double-onboard.test.ts
index c0b7a6021c5..572c4781d4b 100644
--- a/test/e2e/live/double-onboard.test.ts
+++ b/test/e2e/live/double-onboard.test.ts
@@ -458,8 +458,12 @@ test("double-onboard: reuses gateway, preserves sibling sandbox, and recovers st
     "prereq-nemoclaw",
   );
 
+  // OpenShell reaches this fixture from its gateway network namespace, where
+  // the runner's loopback address is not routable.
   const fake = await startFakeOpenAiCompatibleServer({
+    host: "0.0.0.0",
     port: Number(process.env.NEMOCLAW_FAKE_PORT ?? 0),
+    publicHost: "host.openshell.internal",
   });
   await artifacts.writeJson("fake-openai.json", { baseUrl: fake.baseUrl });
   cleanup.trackDisposable("close fake OpenAI-compatible endpoint", async () => {
diff --git a/test/e2e/live/hermes-e2e.test.ts b/test/e2e/live/hermes-e2e.test.ts
index b36d24b18d7..2c2123eb929 100644
--- a/test/e2e/live/hermes-e2e.test.ts
+++ b/test/e2e/live/hermes-e2e.test.ts
@@ -380,7 +380,7 @@ test("hermes-e2e: install.sh onboards Hermes and proves health plus live inferen
 
   if (hermesDashboardE2eEnabled()) {
     expect(resultText(install)).toContain(
-      "Deployment verified — gateway and dashboard are healthy.",
+      "Deployment verified — gateway, dashboard, and inference route are healthy.",
     );
     expect(resultText(install)).toContain("Hermes Agent Dashboard");
     expect(resultText(install)).toContain(`http://127.0.0.1:${HERMES_DASHBOARD_PORT}/`);
diff --git a/test/e2e/live/hermes-inference-switch-helpers.ts b/test/e2e/live/hermes-inference-switch-helpers.ts
index d0e45a86ee3..d81efbecec1 100644
--- a/test/e2e/live/hermes-inference-switch-helpers.ts
+++ b/test/e2e/live/hermes-inference-switch-helpers.ts
@@ -84,16 +84,16 @@ export function mockAnthropicSwitchEnabled(runtimeEnv: NodeJS.ProcessEnv = proce
   );
 }
 
-export function expectAuthenticatedBaselineRequest(
+export function expectAuthenticatedBaselineInventoryRequest(
   baseline: Pick<FakeOpenAiCompatibleServer, "requests"> | undefined,
-  model: string,
 ): void {
   if (!baseline) return;
   expect(baseline.requests()).toContainEqual(
     expect.objectContaining({
       auth: "ok",
-      model,
-      path: "/v1/chat/completions",
+      authorizationSent: true,
+      method: "GET",
+      path: "/v1/models",
     }),
   );
 }
diff --git a/test/e2e/live/hermes-inference-switch.test.ts b/test/e2e/live/hermes-inference-switch.test.ts
index c97dc2fe840..156d9c640c0 100644
--- a/test/e2e/live/hermes-inference-switch.test.ts
+++ b/test/e2e/live/hermes-inference-switch.test.ts
@@ -16,7 +16,7 @@ import {
   ensureCompatibleAnthropicSwitchProvider,
   env,
   envHash,
-  expectAuthenticatedBaselineRequest,
+  expectAuthenticatedBaselineInventoryRequest,
   expectedApiMode,
   expectedBaseUrl,
   hashCheck,
@@ -165,7 +165,7 @@ test("Hermes inference set updates route/config and preserves live runtime", {
 
   const install = await installHermes(host, apiKey, installEnv);
   expect(install.exitCode, resultText(install)).toBe(0);
-  expectAuthenticatedBaselineRequest(mockBaseline, MOCK_BASELINE_MODEL);
+  expectAuthenticatedBaselineInventoryRequest(mockBaseline);
   const baselineRoute = await sandbox.openshell(["inference", "get", "-g", "nemoclaw"], {
     artifactName: "openshell-inference-route-before-switch",
     env: env(),
diff --git a/test/e2e/support/hermes-inference-switch-command-shape.test.ts b/test/e2e/support/hermes-inference-switch-command-shape.test.ts
index 92ad6b5098d..a878ca1f08b 100644
--- a/test/e2e/support/hermes-inference-switch-command-shape.test.ts
+++ b/test/e2e/support/hermes-inference-switch-command-shape.test.ts
@@ -15,6 +15,7 @@ import {
   apiKeyShapeCommand,
   cleanupHermesSwitch,
   compatibleAnthropicMetadataArgs,
+  expectAuthenticatedBaselineInventoryRequest,
   hostedInstallModel,
   inferenceLocalMaxTokens,
   installHermes,
@@ -114,6 +115,36 @@ describe("Hermes inference switch command shape", () => {
     ).toBe("http://host.openshell.internal:18766");
   });
 
+  it("uses authenticated model inventory as baseline readiness evidence", () => {
+    expect(() =>
+      expectAuthenticatedBaselineInventoryRequest({
+        requests: () => [
+          {
+            auth: "ok",
+            authorizationSent: true,
+            bodyBytes: 0,
+            method: "GET",
+            path: "/v1/models",
+          },
+        ],
+      }),
+    ).not.toThrow();
+
+    expect(() =>
+      expectAuthenticatedBaselineInventoryRequest({
+        requests: () => [
+          {
+            auth: "ok",
+            authorizationSent: true,
+            bodyBytes: 0,
+            method: "POST",
+            path: "/v1/models",
+          },
+        ],
+      }),
+    ).toThrow();
+  });
+
   it("enables local baseline inference only for the mock Anthropic lane", () => {
     const mockAnthropic = {
       NEMOCLAW_SWITCH_PROVIDER: "compatible-anthropic-endpoint",