Proof QC agent / Operations

Autonomous proof desk

An agent measures incoming artwork, reads the customer’s instructions and routes the order. A challenger tests the same work in shadow. Code enforces the rules; a person still approves every proof.

Source and eval on GitHub · Today's model spend: …

Simulated order stream

Live desk

Loading fleet state…

Orders come from 106 labelled synthetic cases, not customer traffic. Scheduled ticks run the agents; this page only reads their results.

Incoming orders

Simulation

Connecting to the desk…

  1. Loading orders…

Try your own artwork

Review a sample or upload a file. These manual reviews use the existing single-review flow.

1. The order

Artwork

No artwork chosen yet.

2. The agent's answer

Submit an order to see the decision, the notes and what it cost.

Leaderboard

Saved eval runs on the hand-labelled orders, scored against today's labels. "Rules in code" means order changes are decided by code, not the model.

Recent reviews