We simulate your domain to generate the training data, the evals, and the verifier, then post-train a judge small enough to run on every request.
SimulationTraining dataReal-time judgesPost-training
Dispute management language model
1B params
“There’s an $81.40 charge from Amazon I never made, I want it refunded.”
Mission
How it works
The part we build and operate is the return path. A failure the judge catches in production goes back to the simulator and seeds new cases, which land in the training set and the eval at once.
Simulate
Your domain as an environment. Every case carries the ground truth that built it
Train
SFT on the generator's own inputs and outputs
Verify
The simulator scores the rollout. That is the reward
Serve
Inline at $0.18 per million tokens
Judge
Every action checked against what was asked
Catch
Mismatches held, signed, labeled
Every catch returns to step 01. A real failure seeds new simulations, so one catch becomes a family of cases rather than a single row.
Inside a single request, it is two calls: one between the plan and the action, one after the effect.
The agent
Request
A person states what they want
Plan
The agent selects an action
Act
The action reaches the tool
Effect
Something changes in the world
Zero Proof
Between plan and act. Judge the action against the request, check it is grounded, hold what disagrees.
After the effect. Sign the action and the verdict, and send every catch back to training.
Runs against
Capabilities
Simulation
We build your domain as an environment and generate the cases inside it, covering your failure modes rather than a public benchmark's. The same generator emits the training set, a balanced held-out eval, and the verifier. Collected data gives you inputs and hopes for labels. Simulated data carries the ground truth that produced it, which is what makes RL possible.
Worked example
We built it for refunds and disputes, where a wrong decision costs a chargeback, and published every artifact. Rerun the evaluation against our numbers.
Accuracy against cost to serve
Near-frontier accuracy at about one hundredth of the cost.
Macro intent-type accuracy (%); zeroproof-ecommerce-1b on the full 1,977-row held-out set, frontier models on the balanced comparison subset. Frontier cost at published list prices; Zero Proof cost from measured throughput on a single L4. The dashed line is one pass of post-training on the same base at the same serving cost.
The same method trains zeroproof-ecommerce-0.5b to within five points at half the size.