Run an experiment

Hypothesis — this becomes the first turn of the thread

Bundle

Corpus

Runs

Grader

before you spend anything

Estimated $26.28 and about 41 minutes. At 12 runs the interval lands near ±1.2 points — tight enough to separate from the 92.4% baseline. At 5 runs it would land near ±1.9 and tell you nothing.

Block a keep verdict if autonomy rises while sampled accuracy falls

Those two pull against each other. Every interrupt removed to raise autonomy is a case where the run now guesses instead of asking.