It found the exact branch in our refund code with zero coverage and flagged a fake test that only checked a mock call

## Coverage Gap Analysis — `payments` package (jest, target 85%) Overall line coverage is 61%. The risk is concentrated, not spread — the money paths are the thinnest. | File | Lines | Branches | Verdict | |------|-------|----------|---------| | `refund.service.ts` | 44% | 22% | critical — no error-case tests | | `webhook.verify.ts` | 51% | 0% | critical — signature mismatch path untested | | `cart.utils.ts` | 93% | 88% | healthy | **Top gap:** `refund.service.ts:73-91` — the partial-refund branch and the "amount exceeds capture" guard have zero coverage. A regression here silently over-refunds. ```typescript describe("refundPayment — boundary + error paths", () => { it("rejects a refund larger than the captured amount", async () => { const charge = makeCharge({ capturedCents: 5000 }); await expect(refundPayment(charge, 6000)).rejects.toThrow(/exceeds captured/); }); it("marks the charge partially_refunded on a partial refund", async () => { const charge = makeCharge({ capturedCents: 5000 }); const res = await refundPayment(charge, 2000); expect(res.status).toBe("partially_refunded"); expect(res.refundedCents).toBe(2000); }); }); ``` **Bad test detected:** `refund.test.ts:18` asserts `expect(mock.calledOnce).toBe(true)` — it tests that a method was called, not the resulting balance. That is implementation coupling and gives false confidence. **Plan to 85%:** add the two branch tests above (+9%), cover `webhook.verify.ts` failure path (+12%), then the timeout case in `gateway.ts` (+5%).