A team is comparing two prompt variants for a claims-summary feature over a four-week online test. Both variants call the model through a floating alias that always points to the latest model version. Midway through the test, the alias moves to a newer model. What should the team have done to keep the comparison valid?

Answer: A — Pinned both variants to one dated model snapshot and tested any model upgrade separately

(a) is right: an A/B test should vary only the thing being tested. Pinning a dated snapshot keeps the model fixed, and a model change gets its own offline-then-online evaluation, because prompts tuned for one model don't always transfer. (b) leaves the confound in the data. (c) deliberately confounds the prompt and model changes. (d) is wasteful and still leaves the team exposed to the next alias change.