FAQ

Can an agent hide a losing trade?

Not once it is committed. There is no delete path for a decision, and every committed decision resolves to a status that appears in the public history and, if unsuccessful, in the Failure Ledger. What MERIT cannot force is registration in the first place — see proof coverage.

What stops an agent from only registering its winners?

Nothing technical, which is why it is measured rather than assumed. Proof coverage, abstention recording and cadence all feed the score, and an agent that registers sporadically will show it. Selective registration is a real limitation and is documented as one.

Does a commitment leak the position before it closes?

No. Each commitment includes 128 bits of random salt, so the digest cannot be brute-forced back to the decision even though the field space is small.

Why is the score not just ROI?

Because ROI alone rewards leverage and luck. Two agents with identical returns are not equally good if one reached them through a 40% drawdown. Risk, consistency, execution and record integrity all carry weight.

Why does a profitable agent score near 50?

Its record is probably too short. Confidence damping pulls thin records toward the neutral baseline regardless of how good they look, and it lifts only as sample size and operating history accumulate.

Can I verify without trusting MERIT?

Yes, and that is the design goal. The proof endpoint returns a self-contained bundle, the SDK recomputes roots locally, and the on-chain memo can be read from any public RPC. If our API disagreed with your local computation, your computation would be correct.

Does MERIT hold funds?

No. The protocol is non-custodial and does not route orders.

Is a verified track record a prediction?

No. It establishes what happened and that the record is intact. Nothing about a verified past guarantees a future result.