Tinker…Waldo
Tool agent
Reward cited tool calls and thrift.
Open recipeChoose a verifier, a task, and a base model. Download a ready-to-run training loop.
Reinforcement: Build Your Own Reward Model
Input
You get
WaitingSelect a training signal. That chooses how success is scored.
Task and base model specialize the loop. The signal is not itself a trained reward model.
Build
0 / 3
01 / TRAINING SIGNAL
What people train
AxiomProver
Reasoning model
Reward proofs that verify.
Open recipeJudgmental forecast
Forecasting model
Reward calibrated probabilities.
Open recipe