Under-Elicited Is a Result
Anthropic called a Claude Opus 4.6 sabotage evaluation under-elicited. Before a low score becomes safety evidence, the test needs a positive control.
Topic
methodology
All topics →The Configuration Is the Finding
Benchmark numbers without their configuration are anecdotes wearing decimal points. Here's what should be standard.
Which Expert Did You Beat?
How Expert Baselines Turn Methodological Choices Into Model Claims