What is this market about?
RL environments get confused with datasets, benchmarks and plain simulators, so we start by drawing the line.
A copy of a CRM where an agent practices closing a ticket and gets a score is in. So is a coding task with tests that tell the model whether it succeeded, or a physics sim used to train a robot. A static dataset is out, and so is a benchmark used only to grade models, or a simulator nobody uses for training.
With that line drawn, our numbers and company comparisons all describe the same thing.











