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

The perp-cohort-stats harness publishes perp_venue_funding_24h_bps for every venue and asset it covers: the venue's current funding rate, normalised from its settlement interval (1h, 4h or 8h) to the cost in basis points of holding a long for 24 hours, sampled every minute. The bench averages that series over the trailing 30 days and multiplies by 30: a month of funding at the average daily cost the venue actually quoted. The same figure over 7 days is in the mean column. ETH is the headline; BTC and SOL sit in the p90 and p99 columns on the same convention as the perp-funding bench. The sample count per row is the number of hours behind the average: 720 for a full month, and a venue ranks only with 576 of them (24 days), because a total over fewer days would rank on coverage rather than on cost. Under that floor the row is provisional and unranked, with its day count: a month at the rate of the days measured, which the sample-health badge flags until 15 days and the provisional label until 24.

• ·Source series: perp_venue_funding_24h_bps{venue,asset} from the perp-cohort-stats harness (each venue's own rate endpoint where the harness reads one: Kalshi, Vest, StandX, ApeX, Polymarket, Extended, edgeX, Pacifica, Aster, and Lighter since 2026-09-23, whose window before that date is the Mobula series for the same venue, identical at the switch; the Mobula funding aggregator for Hyperliquid and the CEX rows), sampled every 60 seconds and read here on an hourly grid ([30d:1h]).
• ·30d cost = avg_over_time(series[30d:1h]) x 30: the average daily cost over the hours measured in the window, times a month. A venue is ranked only with 576 hourly samples (24 days) behind that average; under the floor the figure is a month at the rate of the days measured and the row carries its day count. 7d cost is avg_over_time(series[7d:1h]) x 7 (mean column).
• ·Columns: p50 = ETH 30d, p90 = BTC 30d, p99 = SOL 30d, mean = ETH 7d. The column labels are inherited from the latency template; the values are signed bps.
• ·Sign: positive means the long paid funding to shorts; negative means shorts paid longs. Lower ranks first, so on a month where shorts paid, the venue that paid longs the most is first.
• ·Sample health: expected_n is 720 (a month of hourly samples). A venue ranks only with 576 of them (24 days, 80 percent of the grid, which absorbs harness restarts): the figure does not move with coverage, but an average over a few days is not a month's evidence. Under the floor the row is unranked and shows its day count. A feed that stops loses its hours: the harness drops a rate nobody refreshed for five minutes rather than repeat it. Venues added on 2026-09-22 and 23 start under the floor.
• ·Interval normalisation is the venue's: Hyperliquid, Vest, StandX and ApeX settle hourly, Kalshi every 8 hours (three settlements a day), the CEX rows per their own schedule; every row is expressed as cost per 24h before integrating.
• ·This page does not replace perp-funding (bench 036), which ranks the rate right now; it ranks what the last month cost.
• ·A dormant market reads as an expensive one over thirty days. Coinbase International settled its last funding on 1 October 2026 and reports zero notional on 24h, daily and 30 day volume, so the feed kept relaying that final rate: 9.36 bps a day on ETH for twenty of these thirty days, ranking it last at 227 bps against 91 for the next worst. It left the cohort on 21 October. Freshness checks stayed green, because they read a refresh timestamp and a repeated value advances one.