diff --git a/src/__tests__/dashboard-score-banner.unit.test.tsx b/src/__tests__/dashboard-score-banner.unit.test.tsx
index 01bf8ae..d5ff4b6 100644
--- a/src/__tests__/dashboard-score-banner.unit.test.tsx
+++ b/src/__tests__/dashboard-score-banner.unit.test.tsx
@@ -377,7 +377,7 @@ describe('ScoreBanner headline cascade', () => {
     expect(lastFrame()).toContain('3 leaks this period');
   });
 
-  it('headline tier 3 — loops > 100 with cost estimate', () => {
+  it('headline tier 3 — loops > 100, counts and no cost estimate', () => {
     const loops = Array.from({ length: 150 }, () => ({
       toolName: 'Edit',
       commandPreview: '/x',
@@ -399,10 +399,20 @@ describe('ScoreBanner headline cascade', () => {
         filtered={filtered}
       />
     );
-    expect(lastFrame()).toContain('150 loops');
-    expect(lastFrame()).toContain('wasted');
-    // 150 loops * count 10 * COST_PER_LOOP_ITER_USD (0.006) = $9 → "$9.00"
-    expect(lastFrame()).toContain('$9.00');
+    // Was: `150 loops · ~$9.00 wasted`, from Σ(count) × COST_PER_LOOP_ITER_USD.
+    // Wrong three ways — it billed the first legitimate call, counted
+    // long-iteration findings as loops, and used a constant measured ~170x
+    // low. Two inflations and one deflation landed on a believable number
+    // while `node9 scan` reported five times more from the same data.
+    //
+    // The headline now states what it counted. Dollars live on the one screen
+    // that prices them per session.
+    expect(lastFrame()).toContain('150 repeated patterns');
+    expect(lastFrame()).not.toContain('wasted');
+    // Scoped to the headline: the banner has its own spend indicator, so a
+    // bare "no $ anywhere" assertion would fail on unrelated output.
+    const headline = (lastFrame() ?? '').split('📌')[1]?.split('💰')[0] ?? '';
+    expect(headline).not.toContain('$');
   });
 
   it('headline tier 4 — exposed blast paths (when no leaks/loops/early-secrets)', () => {
diff --git a/src/__tests__/report-headline.unit.test.ts b/src/__tests__/report-headline.unit.test.ts
new file mode 100644
index 0000000..b9fe9f7
--- /dev/null
+++ b/src/__tests__/report-headline.unit.test.ts
@@ -0,0 +1,80 @@
+import { describe, it, expect } from 'vitest';
+import { computeHeadline } from '../tui/dashboard/views/report/index.js';
+import { EMPTY_FILTERED_SCAN } from '../tui/dashboard/views/report/derive.js';
+import type { LoopFinding } from '../cli/commands/scan';
+
+// The headline is the first sentence a user reads on the Report screen. It
+// used to price loops itself and got three things wrong at once, landing on a
+// plausible $9.73 while `node9 scan` said $48.66 from the same data.
+
+const loop = (over: Partial<LoopFinding> = {}): LoopFinding => ({
+  toolName: 'Edit',
+  commandPreview: '/tmp/x.ts',
+  count: 50,
+  timestamp: '2026-09-01T00:00:00Z',
+  project: 'p',
+  sessionId: 's1',
+  agent: 'claude',
+  kind: 'loop',
+  ...over,
+});
+
+const ready = { status: 'ready' } as Parameters<typeof computeHeadline>[0];
+
+const scan = (loops: LoopFinding[]) => ({
+  ...EMPTY_FILTERED_SCAN,
+  loops,
+});
+
+describe('computeHeadline — loops', () => {
+  it('reports counts and never a dollar figure', () => {
+    const h = computeHeadline(ready, scan(Array.from({ length: 150 }, () => loop())), null);
+    expect(h?.text).not.toMatch(/\$/);
+  });
+
+  it('separates sustained work from genuinely stuck patterns', () => {
+    // 120 long-iteration + 30 real. The old line called all 150 "loops" and
+    // billed every iteration of all of them.
+    const loops = [
+      ...Array.from({ length: 120 }, () => loop({ kind: 'long-iteration' })),
+      ...Array.from({ length: 30 }, () => loop({ kind: 'loop' })),
+    ];
+    const h = computeHeadline(ready, scan(loops), null);
+    expect(h?.text).toContain('150');
+    expect(h?.text).toContain('30');
+  });
+
+  it('says zero stuck when every finding is sustained work', () => {
+    const loops = Array.from({ length: 150 }, () => loop({ kind: 'long-iteration' }));
+    const h = computeHeadline(ready, scan(loops), null);
+    expect(h?.text).toMatch(/\b0\b/);
+  });
+
+  it('counts a finding with no kind as stuck, matching the rest of the codebase', () => {
+    // `kind` is optional for legacy data; excluding those would under-report,
+    // which is the direction this product keeps failing in.
+    const legacy = loop();
+    delete (legacy as { kind?: unknown }).kind;
+    const h = computeHeadline(ready, scan(Array.from({ length: 150 }, () => legacy)), null);
+    expect(h?.text).toContain('150');
+  });
+
+  it('leaves the >100 threshold alone', () => {
+    const under = computeHeadline(ready, scan(Array.from({ length: 100 }, () => loop())), null);
+    const over = computeHeadline(ready, scan(Array.from({ length: 101 }, () => loop())), null);
+    expect(under?.text ?? '').not.toContain('repeated patterns');
+    expect(over?.text).toContain('repeated patterns');
+  });
+
+  it('still yields to leaks, which outrank loops', () => {
+    const h = computeHeadline(
+      ready,
+      {
+        ...scan(Array.from({ length: 150 }, () => loop())),
+        leaks: [{} as never],
+      },
+      null
+    );
+    expect(h?.text).toContain('leak');
+  });
+});
diff --git a/src/tui/dashboard/views/report/index.tsx b/src/tui/dashboard/views/report/index.tsx
index c9115ad..2f1ff10 100644
--- a/src/tui/dashboard/views/report/index.tsx
+++ b/src/tui/dashboard/views/report/index.tsx
@@ -19,7 +19,6 @@ import { COL } from '../../panels.js';
 import { computeProtection } from '../../data.js';
 import type { BlastSnapshot, ReportPeriod, ScanCache, ShieldStatus } from '../../types.js';
 import type { AggregateResult } from '../../../../cli/aggregate/report-audit.js';
-import { COST_PER_LOOP_ITER_USD } from '@node9/policy-engine';
 
 import { Protection } from './panels/Protection.js';
 import { Cost } from './panels/Cost.js';
@@ -295,7 +294,7 @@ interface Headline {
 
 /** Priority cascade — first match wins. Returns null while there's no
  *  data to assess (e.g. scan still loading and audit/blast both empty). */
-function computeHeadline(
+export function computeHeadline(
   scanCache: ScanCache,
   filtered: FilteredScan,
   blast: BlastSnapshot | null
@@ -326,9 +325,25 @@ function computeHeadline(
     };
   }
   if (filtered.loops.length > 100) {
-    const wasted = filtered.loops.reduce((s, l) => s + (l.count ?? 0) * COST_PER_LOOP_ITER_USD, 0);
+    // Counts, no dollars. This line used to read
+    // `Σ(count) × COST_PER_LOOP_ITER_USD`, which was wrong three ways at once:
+    // it priced the first legitimate call as waste, it counted
+    // `long-iteration` findings (sustained work on one file) as loops, and it
+    // used the flat constant measured at ~170x too low. Two errors inflated
+    // and one deflated, landing on a plausible number — $9.73 against the
+    // $48.66 `node9 scan` reports from the same data. A number that looks
+    // reasonable and is built from three mistakes gets believed.
+    //
+    // Priced correctly needs per-session rates (computeLoopWaste), which this
+    // screen has no access to. Rather than thread that through three layers
+    // for one headline, the money lives on the one screen that computes it
+    // properly. A third dollar figure is a third chance to disagree with
+    // itself.
+    //
+    // "repeated patterns", not "loops": most of these are not loops.
+    const stuck = filtered.loops.filter((l) => l.kind !== 'long-iteration').length;
     return {
-      text: `📌 ${filtered.loops.length} loops · ~${fmtCost(wasted)} wasted`,
+      text: `📌 ${filtered.loops.length} repeated patterns · ${stuck} stuck`,
       color: 'yellow',
     };
   }