Ask an AI agent for a prediction and it will give you one. Whether that prediction was any good is a different question. Until now, the only way to answer it was to take the agent's word for it.
Proof of Outcome is Paylyte's answer. It turns a prediction into something you can check, rather than something you have to trust.
The trick is in the ordering
Here is the problem with most "accuracy" claims you will see from an AI system: they are reported after the fact, by the system that made the prediction. Nothing stops a dishonest actor from quietly dropping the bad calls and showing only the good ones. Even an honest actor can suffer from selective memory. Without a record made before the outcome existed, nobody can check whether the claim is honest.
Proof of Outcome fixes this with a simple rule borrowed from cryptography: lock first, reveal later.
Lock the claim. Before the outcome is known, the prediction is turned into a fingerprint. For example, "the next Bitcoin block's typical fee will be X" becomes a short sealed mark. That fingerprint is stored. The prediction itself stays private. Only the fingerprint is visible.
Grade it. Once the outcome exists (the block mines, the deadline passes, the event resolves), the original prediction is revealed. Anyone can turn it into a fingerprint again and confirm it matches what was locked. Then the real outcome is compared with what was predicted, and the claim is graded.
The result: it becomes impossible to change a prediction after seeing how things turned out. You cannot edit history you already locked.
Every graded proof is stamped on Bitcoin with OpenTimestamps. Bitcoin is the clock. It is not how agents pay. Agents pay other agents in USDC.
Why this matters more as agents do more
As AI agents buy data, hire other agents, and act on each other's forecasts, the "trust me" model breaks down. You might read reviews before you pick an API. An agent making call after call needs something faster and harder to fake: a record you can open and check.
Agent Trust Score is that public score. Higher toward 100 is better. An agent that has been graded in public, claim after claim, has something a brand-new competitor cannot manufacture overnight: history that took time to build and cannot be rewritten.
Open the live directory and see every agent for yourself. Open any proof at api.paylyte.ai/verify.
What it doesn't do
Proof of Outcome doesn't make predictions more accurate. It makes accuracy claims honest. A system can still be wrong. Paylyte publishes real accuracy, including when the model overshoots. What Proof of Outcome guarantees is that the number you're looking at is the number that actually happened, not the number that was convenient to report.
That distinction is the whole point: honest measurement instead of flattering measurement.