Funding paid to hold an ETH perp long for 30 days, per venue, live

The harness records every minute the venue's current funding rate normalised to the cost of holding a long for 24 hours. The bench averages that series over the trailing 30 days on an hourly grid and multiplies by 30: a month of funding at the average daily cost the venue quoted, in basis points of notional. A venue ranks only with 24 days behind that average; with fewer the row shows its day count and no rank. The 7-day figure is the same average over 7 days times 7.

Funding is signed. When the perp trades below spot, shorts pay longs and a long collects funding; the cost is then negative. Lower ranks first, so on such a month the venue that paid longs the most is at the top.

For a long of constant notional held the whole window on a venue with a full month measured, yes, to the precision of a per-minute sample of the quoted rate. A position opened mid-window, resized, or held on a venue that settles funding in discrete 8-hour steps (Kalshi) pays the settlements it was open for; the bench does not model entries and exits. A rate no source refreshed for five minutes is dropped, not repeated, so hours the harness could not read leave the average (at most five minutes of a last value can precede a gap, ten on edgeX and ApeX, which read a ticker cache).

Because the reference a carry trader compares a DEX against is Binance, Bybit or OKX. The CEX rows come from the same Mobula funding feed the perp-funding bench uses (so does Hyperliquid's); the other DEX rows and Kalshi come from each venue's own API.

They joined the cohort on 2026-09-22 (Extended, edgeX, Pacifica, Aster and Lighter RH on 2026-09-23). Their figure is a month at the average daily cost of the days measured since then; the sample count on the row says how many hours are behind it. The badge goes at 15 days (half a month of samples), the provisional label and the rank come at 576 hours, 24 days.