QRef

Claude ports old Qiskit code. QRef runs both versions and shows what actually changed. A person makes the call.

Each page is a static copy of a real local run. Nothing runs on this site. Each page's evidence, manifest and downloadable ZIP are checked against their hashes before publication.

1. Fresh Claude Sonnet port: the circuits are equivalent, and the request for 512 or 17 shots still runs 1,024. The model chose to keep the old behaviour and said so. Its proposal record includes the prompt, notes and code.
2. Earlier research port: the circuits are equivalent, and the port now honours the requested shots. This is a real behaviour change between the two versions.
3. Seeded bit-order regressiondeliberate test: a deliberate, labelled change. The circuit checker still reports equivalence, and QRef's independent classical checks catch the change.
4. Seeded rotation regressiondeliberate test: one mixer angle halved. This time the circuits differ, and the circuit checker (MQT QCEC) reports it. The two seeded cases show which layer catches which kind of change.
5. Positive control: the same port on both sides, including an 8-qubit graph with an isolated node. Result: no differences, exit code 0. This shows that QRef does not invent findings.