diff --git a/src/backend/services/SequentialOrchestrator.ts b/src/backend/services/SequentialOrchestrator.ts
index 8d5848b..eb8f4dc 100644
--- a/src/backend/services/SequentialOrchestrator.ts
+++ b/src/backend/services/SequentialOrchestrator.ts
@@ -91,7 +91,10 @@ export class SequentialOrchestrator<TInput = any, TOutput = any> {
       }
       if (!succeeded) {
         // Phase exhausted all retries. Stop orchestrator.
-        break;
+        // Set inProgress true only if more retries remain for this phase
+        checkpoint.inProgress = (resultsForPhase.length < maxAttempts);
+        await this.saveCheckpoint(checkpoint);
+        return checkpoint;
       }
     }
     checkpoint.inProgress = checkpoint.currentPhase < this.phases.length;
diff --git a/src/backend/tests/orchestrator.test.ts b/src/backend/tests/orchestrator.test.ts
index 1cb31a0..e43a92a 100644
--- a/src/backend/tests/orchestrator.test.ts
+++ b/src/backend/tests/orchestrator.test.ts
@@ -55,21 +55,30 @@ describe('SequentialOrchestrator', () => {
 
   it('persists checkpoint after phase, can resume mid-run after crash', async () => {
     let fired = false;
-    const orchestrator = new SequentialOrchestrator({
-      phases: [
-        { name: 'p1', run: async () => {} },
-        { name: 'p2', run: async () => { if (!fired) {fired=true; throw new Error('fail');} } },
-        { name: 'p3', run: async () => {} },
-      ],
+    // Keep ref to fired in closure used by both orchestrator instances.
+    const phase2Run = async () => { if (!fired) {fired = true; throw new Error('fail');} };
+    const phases = [
+      { name: 'p1', run: async () => {} },
+      { name: 'p2', run: phase2Run, retry: 2 },
+      { name: 'p3', run: async () => {} },
+    ];
+    // First run to fail at p2
+    let orchestrator = new SequentialOrchestrator({
+      phases,
       checkpointPath: tempCheckpoint,
       input: undefined
     });
-    // First run to fail at p2
     let cp = await orchestrator.run();
-    expect(cp.phaseResults.filter(r=>r.phase==='p1' && r.status==='success').length).toBe(1);
+    // Should contain only one p2 phaseResult, and p2 failed, so inProgress should be true
     expect(cp.phaseResults.filter(r=>r.phase==='p2').length).toBe(1);
+    expect(cp.phaseResults.find(r=>r.phase==='p2').status).toBe('failure');
     expect(cp.inProgress).toBe(true);
-    // Simulate restart -- resume should continue from failed phase
+    // Simulate restart -- create NEW orchestrator, resume should continue from failed phase
+    orchestrator = new SequentialOrchestrator({
+      phases,
+      checkpointPath: tempCheckpoint,
+      input: undefined
+    });
     fired = true; // Next attempt will succeed
     cp = await orchestrator.resume();
     expect(cp.phaseResults.filter(r=>r.phase==='p2').length).toBe(2);