An extraction eval stayed green for weeks while users watched their data get mangled. The cause: it graded output against our labels, not against the source.

Our record/replay eval could not run for two days and reported green throughout, because the scorer kept grading cached results from a prompt we had deleted.

Three vcr recording behaviors that never raise an error: the config beats the CLI flag, ambient traffic gets taped, and re-records append instead of replacing.
