Skip to main content
Perps Agent’s “memory” is not a database the team controls. It is an on-chain log of commits and attestations, written to Mantle.

​
Two on-chain artifacts per run

​
1. Commit (before trading)

A hash of the exact strategy config:
  • Venue and market
  • Grid range, step size, order count, sizing
  • Risk caps (max drawdown, max leverage, max position)
  • Timestamp and a run identifier
Written before any order is placed. Once committed, the bot cannot deviate without producing a mismatching attestation.

​
2. Attestation (after trading)

A structured record of the verified outcome:
  • Final realized PnL
  • Max drawdown reached
  • Realized variance / risk-adjusted score
  • Whether the run hit target, stop, or was user-stopped
  • A pointer back to the original commit

​
Why this is “memory”

The AI loop’s “learning” reads this attestation history. Because attestations are public and tied to commits, no one can quietly delete a losing run or invent a winning one to bias the loop.

​
Why this is “proof”

Any observer can:
  1. Pull all commits and attestations for the agent.
  2. Recompute risk-adjusted metrics from the raw attestations.
  3. Verify that each attestation matches its commit.
The track record is independently checkable by anyone, not just shown on a dashboard.
This is the difference between a bot that says “trust me” and one that says “check Mantle.”