Reward Lab. Make success measurable.

Try a simple weighted rubric, change criteria, and see how the computed reward responds. This demo evaluates selected outcomes, not a live model or RL training run.

Reward criteria

01 / DEFINE

Adjust the importance of each rule and simulate pass or fail. Scoring is weighted and normalized to a 0–1 range.

Evaluation result

02 / OBSERVE

WEIGHTED REWARD0.00

Adjust the rubric to calculate a score.

About this score

Real environment rewards must be computed from verified state, not manually checked boxes. This tool is for planning criteria and trade-offs.