How do you design an offline LLM eval harness so the numbers are reproducible and comparable?
The same model can land ten points apart on MMLU depending on prompt format and scoring method. The signal is knowing the knobs (log-prob vs generation, few-shot, normalization) that make evals reproducible.
Updated Sep 2026 · Grounded in real GenAI, LLM, and AI/ML engineering interview loops and written to a senior-engineer editorial bar.
The same model can land ten points apart on MMLU depending on prompt format and scoring method. The signal is knowing the knobs (log-prob vs generation, few-shot, normalization) that make evals reproducible.
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.
