diff --git a/src/policy.js b/src/policy.js
index 2d897c8..84188b0 100644
--- a/src/policy.js
+++ b/src/policy.js
@@ -196,7 +196,7 @@ export function validateImage(policy, image) {
   return normalized;
 }
 
-function validateGatewayImage(image) {
+export function validateGatewayImage(image) {
   const normalized = String(image ?? '').trim();
   if (!isPinnedImage(normalized)) throw new PolicyError('Managed egress requires a sha256 digest-pinned gateway image');
   return normalized;
diff --git a/src/providers/kubernetes.js b/src/providers/kubernetes.js
index 2f16eef..b8550a5 100644
--- a/src/providers/kubernetes.js
+++ b/src/providers/kubernetes.js
@@ -7,7 +7,7 @@ import { commandExists, run, runJson, spawnBackground } from '../process.js';
 import { CleanupError, OwnershipError, ProviderUnavailableError } from '../errors.js';
 import { canonicalResourceRefs, createMetadata, deriveWorkspaceIdentity, labelHash, providerLabels, readCleanupMetadata, readMetadata, workspaceName, WORKSPACE_RUNTIME } from '../metadata.js';
 import { createWorkspaceSecrets, getWorkspaceToken, rotateWorkspaceCredentials, selectGrantedCredentials } from '../auth.js';
-import { requireKubernetesEgress, validateImage } from '../policy.js';
+import { requireKubernetesEgress, validateGatewayImage, validateImage } from '../policy.js';
 import { grantedEgressPolicy } from '../egress-domains.js';
 import { waitForHttpHealth } from '../health.js';
 import { KUBERNETES_TOKEN_FILE, KUBERNETES_TOKEN_MOUNT_PATH, PROVIDER_MODEL_AUTH_FILE, runtimeCommand, runtimeEnvironment } from '../runtime-command.js';
@@ -47,9 +47,19 @@ export function createKubernetesProvider({ policy, sourceDirectory }) {
       }
       throw new ProviderUnavailableError(`Kubernetes cluster is not reachable: ${cause instanceof Error ? cause.message : String(cause)}`, { provider, code: 'WORKSPACE_PROVIDER_CLUSTER_UNREACHABLE', cause });
     });
-    for (const [verb, resource] of requiredPermissions(policy)) {
+    // Each check is a process spawn and a round trip, and there are two dozen of them:
+    // run sequentially they dominated the time to display readiness. Every denial is
+    // collected rather than only the first, so one message names everything to request.
+    const denied = await mapWithConcurrency(requiredPermissions(policy), RBAC_PROBE_CONCURRENCY, async ([verb, resource]) => {
       const { stdout } = await kubectl(['auth', 'can-i', verb, resource, '-n', policy.kubernetes.namespace], { timeoutMs: 20_000 });
-      if (stdout.trim() !== 'yes') throw new ProviderUnavailableError(`Kubernetes RBAC denies ${verb} ${resource} in namespace ${policy.kubernetes.namespace}`, { provider, code: 'WORKSPACE_PROVIDER_RBAC_DENIED' });
+      return stdout.trim() === 'yes' ? null : `${verb} ${resource}`;
+    });
+    const missing = denied.filter(Boolean);
+    if (missing.length > 0) {
+      throw new ProviderUnavailableError(
+        `Kubernetes RBAC denies ${missing.join(', ')} in namespace ${policy.kubernetes.namespace}`,
+        { provider, code: 'WORKSPACE_PROVIDER_RBAC_DENIED' },
+      );
     }
     // Validated after the environment checks because the DNS address is discovered from
     // the cluster, and an unreachable cluster is a more actionable answer than a policy gap.
@@ -71,6 +81,17 @@ export function createKubernetesProvider({ policy, sourceDirectory }) {
     return { provider, available: true, diagnostics: enforcement?.diagnostics ?? [], isolation: enforcement ? { verdict: enforcement.verdict } : null };
   }
 
+/**
+   * The image the isolation probe runs. The probe only needs a runtime that can open a
+   * TCP connection, so it uses the egress gateway image where managed egress already
+   * requires it: a quarter the size of the workspace image, which matters because a
+   * cluster seeing either for the first time must download it before answering.
+   */
+  function isolationProbeImage() {
+    if (policy.egress.mode === 'managed' && policy.egress.gatewayImage) return validateGatewayImage(policy.egress.gatewayImage);
+    return validateImage(policy, policy.defaultImage);
+  }
+
   /**
    * The address of the cluster's DNS service, which is different on every cluster and is
    * therefore discovered rather than asked for. The setting remains an override, and is
@@ -138,8 +159,8 @@ export function createKubernetesProvider({ policy, sourceDirectory }) {
       return { action, namespace, created: !exists };
     }
     if (action === 'check-isolation') {
-      const result = await checkNetworkPolicyEnforcement(kubectl, { context: policy.kubernetes.context, namespace: policy.kubernetes.namespace, image: validateImage(policy, policy.defaultImage), force: true });
-      return { action, verdict: result.verdict, diagnostics: result.diagnostics };
+      const result = await checkNetworkPolicyEnforcement(kubectl, { context: policy.kubernetes.context, namespace: policy.kubernetes.namespace, image: isolationProbeImage(), force: true });
+      return { action, verdict: result.verdict, diagnostics: result.diagnostics, imageUnavailable: result.imageUnavailable === true };
     }
     throw new Error(`Unsupported Kubernetes setup action: ${action}`);
   }
@@ -156,7 +177,7 @@ export function createKubernetesProvider({ policy, sourceDirectory }) {
     // Every isolation guarantee this provider makes rests on the cluster enforcing the
     // NetworkPolicies it writes, and acceptance of the objects proves nothing. Verified
     // here rather than in preflight so listing and readiness stay cheap.
-    const enforcement = await requireNetworkPolicyEnforcement(kubectl, { provider, context: policy.kubernetes.context, namespace: policy.kubernetes.namespace, image: validateImage(policy, policy.defaultImage) });
+    const enforcement = await requireNetworkPolicyEnforcement(kubectl, { provider, context: policy.kubernetes.context, namespace: policy.kubernetes.namespace, image: isolationProbeImage() });
     const meta = readMetadata(info, provider, policy);
     const identity = identityFromMetadata(meta);
     const refs = canonicalResourceRefs(meta.providerResourceID, provider, policy);
@@ -593,6 +614,23 @@ function portAvailable(port) {
   return new Promise((resolve) => { const server = createServer(); server.once('error', () => resolve(false)); server.listen(port, '127.0.0.1', () => server.close(() => resolve(true))); });
 }
 
+const RBAC_PROBE_CONCURRENCY = 8;
+
+/** Runs `task` over `items` with a bounded number in flight, preserving input order. */
+async function mapWithConcurrency(items, limit, task) {
+  const results = new Array(items.length);
+  let next = 0;
+  const workers = Array.from({ length: Math.min(limit, items.length) }, async () => {
+    for (;;) {
+      const index = next++;
+      if (index >= items.length) return;
+      results[index] = await task(items[index], index);
+    }
+  });
+  await Promise.all(workers);
+  return results;
+}
+
 function requiredPermissions(policy) {
   const resources = ['pods', 'secrets', 'serviceaccounts', 'persistentvolumeclaims', 'services', 'deployments.apps', 'networkpolicies.networking.k8s.io'];
   const result = resources.flatMap((resource) => ['create', 'get', 'delete'].map((verb) => [verb, resource]));
diff --git a/src/providers/network-policy-enforcement.js b/src/providers/network-policy-enforcement.js
index 2c79758..3c9cda1 100644
--- a/src/providers/network-policy-enforcement.js
+++ b/src/providers/network-policy-enforcement.js
@@ -122,7 +122,12 @@ async function runProbePod(kubectl, manifest, namespace, deadline, now) {
     const waiting = container?.state?.waiting;
     // An unpullable image never terminates, so treat it as a probe failure rather than
     // waiting out the whole deadline for a pod that cannot run.
-    if (waiting && /ErrImagePull|ImagePullBackOff|InvalidImageName|CreateContainerConfigError/i.test(waiting.reason ?? '')) {
+    if (waiting && /ErrImagePull|ImagePullBackOff|InvalidImageName/i.test(waiting.reason ?? '')) {
+      // Distinguished from other start failures because the cause is a setting the
+      // operator owns, not anything about the cluster's networking.
+      return { blocked: `the workspace image could not be pulled by the cluster`, imageUnavailable: true };
+    }
+    if (waiting && /CreateContainerConfigError/i.test(waiting.reason ?? '')) {
       return { blocked: `probe pod ${name} cannot start: ${waiting.reason}${waiting.message ? ` (${waiting.message})` : ''}` };
     }
     if (status.phase === 'Failed' && !terminated) return { blocked: `probe pod ${name} failed: ${status.reason ?? 'unknown reason'}` };
@@ -144,7 +149,7 @@ export async function probeNetworkPolicyEnforcement(kubectl, { namespace, image,
   try {
     const baseline = await runProbePod(kubectl, probePod(`${baseName}-baseline`, namespace, { ...labels, 'openchamber.io/probe-role': 'baseline' }, image, 'baseline'), namespace, deadline, now);
     created.push(['pod', `${baseName}-baseline`]);
-    if (baseline.blocked) return { verdict: ENFORCEMENT_VERDICTS.INCONCLUSIVE, diagnostics: [`Network isolation could not be verified: ${baseline.blocked}.`] };
+    if (baseline.blocked) return { verdict: ENFORCEMENT_VERDICTS.INCONCLUSIVE, diagnostics: [`Network isolation could not be verified: ${baseline.blocked}.`], imageUnavailable: baseline.imageUnavailable === true };
     if (baseline.exitCode !== 0) {
       return {
         verdict: ENFORCEMENT_VERDICTS.INCONCLUSIVE,
@@ -156,7 +161,7 @@ export async function probeNetworkPolicyEnforcement(kubectl, { namespace, image,
     created.push(['networkpolicy', baseName]);
     const restricted = await runProbePod(kubectl, probePod(`${baseName}-restricted`, namespace, restrictedLabels, image, 'restricted'), namespace, deadline, now);
     created.push(['pod', `${baseName}-restricted`]);
-    if (restricted.blocked) return { verdict: ENFORCEMENT_VERDICTS.INCONCLUSIVE, diagnostics: [`Network isolation could not be verified: ${restricted.blocked}.`] };
+    if (restricted.blocked) return { verdict: ENFORCEMENT_VERDICTS.INCONCLUSIVE, diagnostics: [`Network isolation could not be verified: ${restricted.blocked}.`], imageUnavailable: restricted.imageUnavailable === true };
     if (restricted.exitCode === 0) return { verdict: ENFORCEMENT_VERDICTS.NOT_ENFORCED, diagnostics: [] };
     return { verdict: ENFORCEMENT_VERDICTS.ENFORCED, diagnostics: [] };
   } finally {
diff --git a/src/providers/network-policy-enforcement.test.js b/src/providers/network-policy-enforcement.test.js
index d01da2b..fc33583 100644
--- a/src/providers/network-policy-enforcement.test.js
+++ b/src/providers/network-policy-enforcement.test.js
@@ -69,7 +69,10 @@ describe('kubernetes network policy enforcement probe', () => {
     const result = await probeNetworkPolicyEnforcement(kubectl, { namespace: 'workspaces', image: IMAGE });
 
     expect(result.verdict).toBe(ENFORCEMENT_VERDICTS.INCONCLUSIVE);
-    expect(result.diagnostics[0]).toMatch(/ImagePullBackOff/);
+    // The cause is a setting the operator owns, so it is reported apart from anything
+    // the probe learned about the cluster's networking.
+    expect(result.imageUnavailable).toBe(true);
+    expect(result.diagnostics[0]).toMatch(/image could not be pulled/i);
   });
 
   it('removes every probe resource it created, including after a verdict of no enforcement', async () => {