Agent LP with Verifiable Yield — an autonomous concentrated-liquidity agent paired with a deterministic replay harness that makes LP performance publicly checkable. The agent manages positions on Uniswap V3-compatible pools; the harness reproduces its numbers from on-chain data and calibrates them against rewards actually claimed by live positions (calibration ratios: 1.0066, 0.9924).
Make LP and agent yield claims independently verifiable: any agent's claimed performance becomes checkable by third parties through deterministic replay + live calibration, instead of declarative trust.
LP yield claims are opaque: quoted APYs ignore re-range costs, exit timing, rewardless stretches and tail risk. There is no standard, reproducible way for a third party to verify whether a claimed LP strategy number actually reproduces from on-chain data. Grant evaluators, treasury managers and other agents must trust prose.
A deterministic, open-source measurement harness (pure Python, seeded, no LLM) that replays LP strategies over collected on-chain snapshots (946+ hours of hourly sqrtPrice/rewardGrowth data) and calibrates model output against rewards actually claimed by live positions. Statistics: 113 seeded entry points, 14/30-day horizons, medians with p10/p90, tail probabilities, costs included. Every claimed reward and position change is in a public journal. The same format runs on any EVM chain — Celo port is the next deliverable.
