Skip to content

How-to recipes

One task per page, each with the commands and the output they really produce. Beginners are better served by the tutorial ladder; these pages assume the vocabulary and get to the point.

I want to… Recipe
pick or compose the reward a run is scored under Choose a reward
write the schedule that shifts a run's conditions Write a schedule
score a trace I produced somewhere else Score a run you already have
change what the metrics measure against Tune the protocol
understand the notes under a comparison table Read the aggregate's flags
make two runs of one document come out identical Make a run reproducible
turn a trained run into a servable artifact Export and serve a policy
check my environment, reward, algorithm or provider Check your plugin
compare algorithm families across seeds tutorial L4 owns this one end to end