TLDC has built an open-source autograder for rubric-based reinforcement learning and evaluation. Below we grade a few samples from Perplexity’s DRACO, a benchmark we co-created with Perplexity using our rubric generation software. For each finance question from DRACO, you can see four outputs being graded with rubric.

Watch all four outputs get scored to see a complete GRPO advantage calculation. Hover over outputs or criteria to see full text.