A customer says 'the model got worse, and we changed nothing.' What are your hypotheses, and how do you test each?
The classic FDE escalation. Six hypotheses, silent model updates, input drift, prompt entropy, sampling noise, perception shift, and the eval-replay test that splits them in an afternoon.
Updated Aug 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.
The classic FDE escalation. Six hypotheses, silent model updates, input drift, prompt entropy, sampling noise, perception shift, and the eval-replay test that splits them in an afternoon.
Lead with where the obvious approach breaks, because that is the judgment they are screening for — most candidates jump straight to the happy path and lose the room.
Then walk the failure back through the pipeline in order, naming the one metric the customer's exec sponsor actually cares about before you propose the fix.
