Test 3: Verifying empty cases (no false positives)...
Found 5 ground truth file(s)

❌ 2.json: False positive detected!
    Ground truth: 0 items
    Output: 3 items
    The LLM generated action items when there should be none for Michal
✓ 3.json: Correctly empty (no false positives)
❌ 5.json: Failed to load JSON: Extra data: line 1 column 133 (char 132)

============================================================
Results: 1/2 empty cases correct
============================================================
FAILED: Some files have false positives (generated items when there should be none)