diff --git a/src/__tests__/wizard-abort.test.ts b/src/__tests__/wizard-abort.test.ts
index 8a72e0cf..b35bc2c7 100644
--- a/src/__tests__/wizard-abort.test.ts
+++ b/src/__tests__/wizard-abort.test.ts
@@ -2,6 +2,7 @@
 import {
   wizardAbort,
   WizardError,
+  abortFingerprint,
   registerCleanup,
   clearCleanup,
   runCleanups,
@@ -108,7 +109,9 @@ describe('wizardAbort', () => {
 
     await expect(wizardAbort({ error })).rejects.toThrow('process.exit called');
 
-    expect(mockAnalytics.captureException).toHaveBeenCalledWith(error, {});
+    expect(mockAnalytics.captureException).toHaveBeenCalledWith(error, {
+      $exception_fingerprint: 'wizard_abort_something_broke',
+    });
     expect(mockAnalytics.shutdown).toHaveBeenCalledWith('error');
   });
 
@@ -127,11 +130,24 @@ describe('wizardAbort', () => {
     await expect(wizardAbort({ error })).rejects.toThrow('process.exit called');
 
     expect(mockAnalytics.captureException).toHaveBeenCalledWith(error, {
+      $exception_fingerprint: 'wizard_abort_mcp_missing',
       integration: 'nextjs',
       error_type: 'MCP_MISSING',
     });
   });
 
+  it('lets a WizardError context override the default fingerprint', async () => {
+    const error = new WizardError('OAuth error: timeout', {
+      $exception_fingerprint: 'wizard_oauth_timeout',
+    });
+
+    await expect(wizardAbort({ error })).rejects.toThrow('process.exit called');
+
+    expect(mockAnalytics.captureException).toHaveBeenCalledWith(error, {
+      $exception_fingerprint: 'wizard_oauth_timeout',
+    });
+  });
+
   it('runs registered cleanup functions before analytics and display', async () => {
     const callOrder: string[] = [];
 
@@ -178,6 +194,56 @@ describe('wizardAbort', () => {
   });
 });
 
+describe('abortFingerprint', () => {
+  it('is identical for the same cause raised from different install paths', () => {
+    const fromNpx = new WizardError(
+      'orchestrator drain ended with failed tasks',
+    );
+    fromNpx.stack =
+      'WizardError\n    at /home/u/.npm/_npx/2f0a/node_modules/@posthog/wizard/dist/index.js:1:1';
+    const fromPnpmDlx = new WizardError(
+      'orchestrator drain ended with failed tasks',
+    );
+    fromPnpmDlx.stack =
+      'WizardError\n    at /root/.cache/pnpm/dlx/9b7c1f/node_modules/@posthog/wizard/dist/index.js:1:1';
+
+    expect(abortFingerprint(fromNpx)).toBe(abortFingerprint(fromPnpmDlx));
+    expect(abortFingerprint(fromNpx)).toBe(
+      'wizard_abort_orchestrator_drain_ended_with_failed_tasks',
+    );
+  });
+
+  it('separates the failed-tasks guard from the never-ran guard', () => {
+    expect(
+      abortFingerprint(
+        new WizardError('orchestrator drain ended with failed tasks'),
+      ),
+    ).not.toBe(
+      abortFingerprint(
+        new WizardError('orchestrator drain ended with tasks that never ran'),
+      ),
+    );
+  });
+
+  it('truncates long messages so a shared cause stays one group', () => {
+    const suffix =
+      'API Error: The socket connection was closed unexpectedly. For more information, pass `verbose: true`';
+    const once = new WizardError(`API error: ${suffix}`);
+    const twice = new WizardError(`API error: ${suffix}\n${suffix}`);
+
+    expect(abortFingerprint(once)).toBe(abortFingerprint(twice));
+    // Prefix + 80 chars of slug, and never a trailing separator.
+    expect(abortFingerprint(once).length).toBeLessThanOrEqual(
+      'wizard_abort_'.length + 80,
+    );
+    expect(abortFingerprint(once)).not.toMatch(/_$/);
+  });
+
+  it('falls back to the error name when there is no message', () => {
+    expect(abortFingerprint(new TypeError(''))).toBe('wizard_abort_typeerror');
+  });
+});
+
 describe('runCleanups', () => {
   beforeEach(() => {
     clearCleanup();
diff --git a/src/lib/agent/runner/sequence/orchestrator/__tests__/drain-failure.test.ts b/src/lib/agent/runner/sequence/orchestrator/__tests__/drain-failure.test.ts
new file mode 100644
index 00000000..f6fe205a
--- /dev/null
+++ b/src/lib/agent/runner/sequence/orchestrator/__tests__/drain-failure.test.ts
@@ -0,0 +1,45 @@
+import { drainFailure } from '../drain-failure';
+
+describe('drainFailure', () => {
+  it('returns nothing when the drain finished everything', () => {
+    expect(
+      drainFailure({ failed: 0, blocked: 0, failedTypes: '' }),
+    ).toBeUndefined();
+  });
+
+  it('reports failed tasks when retries were exhausted', () => {
+    expect(
+      drainFailure({ failed: 1, blocked: 0, failedTypes: 'instrument' }),
+    ).toEqual({
+      error: 'orchestrator drain ended with failed tasks',
+      reason: 'the instrument step failed',
+    });
+  });
+
+  it('does not call a blocked-only drain a failure', () => {
+    const failure = drainFailure({ failed: 0, blocked: 3, failedTypes: '' });
+
+    expect(failure?.error).toBe(
+      'orchestrator drain ended with tasks that never ran',
+    );
+    expect(failure?.error).not.toContain('failed');
+    expect(failure?.reason).toBe('3 steps never ran');
+  });
+
+  it('keeps the reason grammatical for a single blocked task', () => {
+    expect(
+      drainFailure({ failed: 0, blocked: 1, failedTypes: '' })?.reason,
+    ).toBe('1 step never ran');
+  });
+
+  it('prefers the failure when tasks both failed and stayed blocked', () => {
+    // Blocked tasks are usually downstream of the failure, so the failed types
+    // are the actionable detail.
+    expect(
+      drainFailure({ failed: 1, blocked: 2, failedTypes: 'review' }),
+    ).toEqual({
+      error: 'orchestrator drain ended with failed tasks',
+      reason: 'the review step failed',
+    });
+  });
+});
diff --git a/src/lib/agent/runner/sequence/orchestrator/drain-failure.ts b/src/lib/agent/runner/sequence/orchestrator/drain-failure.ts
new file mode 100644
index 00000000..2c39fd7e
--- /dev/null
+++ b/src/lib/agent/runner/sequence/orchestrator/drain-failure.ts
@@ -0,0 +1,38 @@
+/**
+ * Which guard, if any, should fail an orchestrator run once the drain returns.
+ *
+ * A drain can end short two ways, and they are different defects: a task
+ * exhausted its retries (`Failed`), or a task never ran at all because a
+ * dependency never completed (still `Pending` with nothing runnable). Both mean
+ * the wizard did not set PostHog up, so both abort — but the exception has to
+ * name the one that happened. A blocked-only drain has nothing in `Failed`, so
+ * reporting it as "failed tasks" sent triage hunting for a failure the queue
+ * state does not contain.
+ */
+export type DrainFailure = {
+  /** Stable exception message — also the error-tracking group key. */
+  error: string;
+  /** The one-liner the user sees, minus the contact line. */
+  reason: string;
+};
+
+export function drainFailure(args: {
+  failed: number;
+  blocked: number;
+  /** Comma-joined types of the tasks in `Failed`, for the user-facing reason. */
+  failedTypes: string;
+}): DrainFailure | undefined {
+  if (args.failed > 0) {
+    return {
+      error: 'orchestrator drain ended with failed tasks',
+      reason: `the ${args.failedTypes} step failed`,
+    };
+  }
+  if (args.blocked > 0) {
+    return {
+      error: 'orchestrator drain ended with tasks that never ran',
+      reason: `${args.blocked} step${args.blocked === 1 ? '' : 's'} never ran`,
+    };
+  }
+  return undefined;
+}
diff --git a/src/lib/agent/runner/sequence/orchestrator/orchestrator-runner.ts b/src/lib/agent/runner/sequence/orchestrator/orchestrator-runner.ts
index 3b623ef8..196339b6 100644
--- a/src/lib/agent/runner/sequence/orchestrator/orchestrator-runner.ts
+++ b/src/lib/agent/runner/sequence/orchestrator/orchestrator-runner.ts
@@ -54,6 +54,7 @@ import {
   type QueuedTask,
 } from './queue';
 import { drainQueue, type RunTask } from './executor';
+import { drainFailure } from './drain-failure';
 import { RunMetrics } from './run-metrics';
 import {
   agentRunTools,
@@ -620,22 +621,24 @@ export async function runOrchestrator(
   const notRequired = summary[TaskStatus.Skipped];
 
   // A drain that ends with failed tasks (retries exhausted) or tasks still
-  // pending (blocked behind a failed dependency) did NOT set PostHog up —
-  // abort like a linear agent failure instead of claiming success.
+  // pending (blocked behind a dependency that never completed) did NOT set
+  // PostHog up — abort like a linear agent failure instead of claiming success.
+  // `drainFailure` picks which of the two it was so the exception says so.
   const blocked = summary[TaskStatus.Pending];
-  if (summary.failed > 0 || blocked > 0) {
-    const failedTypes = store
-      .list()
-      .filter((t) => t.status === TaskStatus.Failed)
-      .map((t) => t.type)
-      .join(', ');
+  const failedTypes = store
+    .list()
+    .filter((t) => t.status === TaskStatus.Failed)
+    .map((t) => t.type)
+    .join(', ');
+  const failure = drainFailure({
+    failed: summary.failed,
+    blocked,
+    failedTypes,
+  });
+  if (failure) {
     await wizardAbort({
-      message: `The wizard was unable to set up PostHog: ${
-        failedTypes
-          ? `the ${failedTypes} step failed`
-          : `${blocked} steps never ran`
-      }.\n\nPlease report this to: ${WIZARD_CONTACT_EMAIL}`,
-      error: new WizardError('orchestrator drain ended with failed tasks', {
+      message: `The wizard was unable to set up PostHog: ${failure.reason}.\n\nPlease report this to: ${WIZARD_CONTACT_EMAIL}`,
+      error: new WizardError(failure.error, {
         tasks_failed: summary.failed,
         tasks_blocked: blocked,
         failed_types: failedTypes,
diff --git a/src/utils/wizard-abort.ts b/src/utils/wizard-abort.ts
index fe844728..94339c07 100644
--- a/src/utils/wizard-abort.ts
+++ b/src/utils/wizard-abort.ts
@@ -29,6 +29,34 @@ interface WizardAbortOptions {
   exitCode?: number;
 }
 
+/**
+ * Stable error-tracking group key for an abort.
+ *
+ * Error tracking fingerprints server-side from the stack trace, and every
+ * install path is different — pnpm dlx content hashes, `.npm/_npx` dirs, CI
+ * checkouts, local clones. So one cause shatters into an issue per install,
+ * each reading "1 occurrence, 1 user, 1 session", and the real recurrence rate
+ * becomes unmeasurable. Aborts are raised from a fixed set of guard sites, so
+ * the error's own message is the stable identity; supplying it collapses those
+ * variants into one issue.
+ *
+ * Truncation is deliberate: it folds together messages that share a cause but
+ * differ in their tail (a repeated API error suffix, an inlined quota URL).
+ *
+ * A callsite that wants different grouping sets `$exception_fingerprint` in the
+ * WizardError context — that wins over this default.
+ */
+export function abortFingerprint(error: Error): string {
+  const slug = (value: string) =>
+    value
+      .toLowerCase()
+      .replace(/[^a-z0-9]+/g, '_')
+      .replace(/^_+|_+$/g, '')
+      .slice(0, 80)
+      .replace(/_+$/, '');
+  return `wizard_abort_${slug(error.message) || slug(error.name) || 'unknown'}`;
+}
+
 const cleanupFns: Array<() => void> = [];
 
 export function registerCleanup(fn: () => void): void {
@@ -72,6 +100,7 @@ export async function wizardAbort(
   // 2. Capture error in analytics (if provided)
   if (error) {
     analytics.captureException(error, {
+      $exception_fingerprint: abortFingerprint(error),
       ...((error instanceof WizardError && error.context) || {}),
     });
   }