How long does a “best yield” answer actually stay true? This page measures it on 452 timestamped attestations published to Base mainnet over 24.75 days — the service's own past calls, scored against what happened next.
| Market | Leadership changes | Median time on top | Lasted ≥24h | Round-trip share | Attestations |
|---|---|---|---|---|---|
WETH — WETH lending (Base) | 0 | never changed | — | — | 81 |
ETH_STAKING — ETH liquid staking (Ethereum) | 10 | 26h | 70% | 50% | 81 |
USDC — USDC lending (Base) | 173 | 2h | 2% | 48% | 290 |
Round-trip share is how much of the rotation is the same two protocols trading places — churn rather than genuine change. A market where the leader flips every couple of hours, half the time back to where it came from, is not offering an edge worth paying gas for.
No. the size of the lead does NOT predict how long it lasts: a wide gap was no more durable than a narrow one. Do not use gap size as a proxy for reliability. Measured across 183 completed leadership spells (Spearman -0.0172).
| Lead at the start of the spell | Spells | Median duration |
|---|---|---|
0-24bps | 68 | 2h |
25-99bps | 62 | 2h |
100-299bps | 39 | 2h |
>=300bps | 14 | 2h |
This one is published because it is inconvenient: a wide gap reads like a confident signal, and the data says it is not more durable than a narrow one. Anyone using gap size as a proxy for reliability — including anyone using this service — is using a proxy that does not hold.
Every input is a public attestation on Base mainnet. Query https://base.easscan.org/graphql for attester 0x561143BFE9E2D975D92e915B8EfFEAa54119472a and schema 0xe74a27f6c216134a1a3aef4c26e29bd8866ac679a8023ddde34faa0bb05dd272, group consecutive attestations by leader per asset, and you get these numbers. Query this schema and attester, group consecutive attestations by bestProtocol per asset, and every figure above follows. Nothing here depends on trusting this server. Machine-readable at /persistence.json.
Honest limits, stated here rather than buried: the observation interval is 1h, so leads shorter than that cannot be distinguished; a market whose leader has not changed reports a floor, never a median; and every figure carries its sample size, with “—” wherever the sample is too small to support a number.
Which protocol leads right now, how long it has already led, and the lead duration applied to MOVE/HOLD decisions are the paid part. /persistence/eth-staking-yield, /persistence/usdc-base-yield, /persistence/weth-base-yield — priced in the 402 challenge, payable per call over x402 with no account. The history above stays free: it is the evidence, and evidence behind a paywall proves nothing.