Golden-run evidence
Golden-run evidence is the saved, checkable record of a test run: the inputs, the outputs, the pass/fail results, and the timestamp, stored somewhere you can point to later. It turns "the tests passed" into a file you can open and inspect. The catch: it only counts if it was captured automatically during the real run. Type it up afterward from memory and it proves nothing.
Why it matters
A green checkmark in a chat window proves nothing once the window is closed. When a client, a teammate, or future-you asks "did this actually pass?", a claim is worthless and a stored run settles it. Without golden-run evidence you re-run everything to answer basic questions, or you trust a screenshot that might be stale or cherry-picked. It is the difference between saying it works and being able to show it works.
How it works
Your CI job writes the run to a durable artifact: a JSON or JUnit file with each case, its status, and the commit hash it ran against. Good evidence is tamper-evident and tied to a specific version, so you know exactly what code produced it. Store it next to the commit or in a build log, and make the pass/fail gate read from that file instead of a human summary. If you cannot open the record and see the individual results, you do not have evidence. You have a vibe.
A client asks whether your refund-policy bot still answers the 14-day window correctly after last week's prompt change. Instead of re-running it live and hoping, you open the golden-run artifact from that deploy: 87 cases, all green, tied to commit a3f9c2, timestamped Tuesday. You send the file, and the question is closed in one message.