How many bitcoin worth of perpetual bets (bets on price with no expiry) are open on each of five exchanges, at the end of each day.
93,677 BTC By exchange · BTC · Binance on Oct-25 2026, the last day in this build, measured and published by askthetape.com.
A perpetual bet stays open until someone closes it. Each line is the number of bitcoin worth of such bets open on one exchange at the end of the day. We show whether open positions increased or decreased; this count does not tell us why they closed. The Binance line reaches back to 2020-09-01 because the exchange publishes its own archive of the same snapshots, and we read the last one of each day from it — the same rule we apply to our own capture. The other four start on 2026-03-31, where our capture did; the mark on the chart is the day the archive hands over to us. Historical archive fetched 2026-10-06 from the venue's public endpoint / the public CFTC endpoint; source units and gaps preserved; browser-only.
| leg | own OI snapshots per exchange and coin, last of the UTC day, in coin units |
|---|---|
| cadence | snapshot by snapshot, daily view |
| since | Binance 2020-09-01, from the exchange's own public archive — Binance's public daily archives; the other four 2026-03-31, from our own snapshots. The archive leg stops the day before ours starts, so no day is counted twice |
| leg · Binance before our capture | Binance's public daily archives — the exchange's own daily archive of perpetual metrics for BTCUSDT, 5-minute snapshots, 2020-09-01 → 2026-03-30. We take the last NON-ZERO snapshot of each UTC day — that is the rule, stated exactly: on 5 days the archive's own tail is zeros and the close falls earlier (2021-02-19, 2021-12-30, 2022-03-07, 2022-06-24, 2024-02-16), because a zero is not an open interest. The archive's unit is the same one our leg uses: contracts of the base asset, i.e. BTC, one side. 2030 days sit ahead of our first day; 2101 of the 2197 archive days carry the full 288 snapshots. 🩸 the archive has no monthly bundle and no day before 2020-09-01, so the line starts there and not at the contract's 2019 listing. 🩸 7 day(s) of the archive stretch are absent from the line: those are days the archive ends on a zero snapshot (a hole of the exchange's own, not ours), and a zero is not an open interest. |
| 🩸 days the archive left short | 96 of the days it carries arrive with fewer than the full 288 five-minute snapshots. A day thinner than half a day is not published at all (that removed 5 days: 2021-06-22, 2022-05-27, 2022-06-27, 2022-09-03, 2022-12-14, the worst of them 12 snapshots closing at 07:55); the rest stay, and the snapshot count travels with every point, so a thin day can be seen rather than guessed. |
| check · the same archive, read a second time | our own production poller keeps the same Binance metrics since the same 2020-09-01 (2194 days overlap this leg). It is NOT an independent instrument — it reads the same archive, only it closes the day on an earlier snapshot — its 23:00 one (probed on 40 dates spread over 2020-09 → 2026-09: 38 of them match 23:00 exactly, and the two that match 22:55 are days the archive itself was later revised by one step), while this page closes it on the last non-zero one. The two differ by a median of 0.2521 % (p95 1.4730 %; 235 days apart by more than 1 %, and 3 days where the poller's file carries a zero — those zero days are left out of this comparison rather than counted as a 100 % gap). Same instrument, one hour apart at the end of the day. |
| check · where archive and our own capture overlap | both carry 167 days. Regression through zero of archive on ours: 0.999988; median |Δ| 0.0157 % (worst 0.451 %) — measured on THIS build, not quoted. The same 167 days tested against the archive's day-MEAN instead of its last snapshot give 0.8576 % — 55 times worse, which is why the splice uses the last snapshot: a mean-spliced history would print a different number than this page's own rule. |
| calibration | 🩸 ONE SIDE (Bybit's own singleOpenInterest) EVERYWHERE — one convention across the whole site, not per page. Bybit and Hyperliquid count BOTH sides of every contract in their public field, so both are halved. Bybit: checked again on the live handles 2026-09-12, openInterest / singleOpenInterest = 2.000000 on all seven symbols probed (BTC, SOL, HYPE, UNI and the TradFi books XAUT, SPX, TSLA — the exchange's convention, not one market's). Hyperliquid reports two-sided open interest — its own note of 2026-09-23: «Two-sided OI is calculated as the sum of long and short positions» — and our own record of its positions agrees: on no market do the tracked longs reach much past half of the field. We halve it to one side, like the other venues, while Binance 103,753.375, OKX 27,373.52 and Bitget 35,260.41 already publish one side and are left alone. Our own capture of the same hour sits within 0.1-0.2% of each one-sided handle and at 2.0014x of Bybit's single side. The correction runs over the WHOLE history — neither field has ever been one-sided, so there is no seam to date; until this correction, Hyperliquid was printed here at both sides. Same figure day by day as «How much is bet, on every venue»: measured on this build, the unit-free Bybit ÷ Binance ratio agrees between the two pages to within 0.02% (median 1.0001 over the days both carry, range 0.9999-1.0002), which is the test a ×2 mismatch could not pass. 🩸 Bybit's own daily archive, which that page uses before our capture starts (2026-03-30 and earlier), stamps each bar 00:00Z — so the bar labelled D is the CLOSE of D-1: measured 2026-09-12 against our own capture, our day D halved equals their day D+1 with median ratio 0.999973 over 164 common days (5th-95th percentile 0.9990-1.0011), while the same-day pairing is only 1.0011 with a 0.92-1.08 spread. That one-day stamp is the archive's, not ours, and it is named there too. |
| calibration · how the factor was measured | 🩸 one side per venue. Bybit's endpoint reports BOTH sides and is halved here: measured against Bybit's own archive over 822 venue-coin-days (median ratio 2.0000, from the calibration that ships with the build) on a window of 209 days per coin, and against the exchange's second handle in one moment (openInterest / singleOpenInterest = 2.000000, 2026-09-11). Binance 1.0002 and OKX 1.00 need no correction. Every venue on this page counts one side, so the five lines can be compared with each other and with «How much is bet, on every venue». 🩸 Hyperliquid counts both sides too — its own note of 2026-09-23: «Two-sided OI is calculated as the sum of long and short positions» — so it is halved here as well, over its whole history: our prod capture of it reads 2.0000 times our own five-minute capture of the same venue brought to one side (median over 184 common venue-coin-days). Until this correction its line here was twice one side. |
| archival source | Historical archive fetched 2026-10-06 from the venue's public endpoint / the public CFTC endpoint; source units and gaps preserved; browser-only. |
| historical views | Browser-only historical views retain the source dates and units. |
| this build | data to Oct-25 2026 |
| policy | no backfill · read it here · export: from 2026-03-31 (rows before that day are the exchange's own archive and stay here) |
The chart itself is drawn in the browser: turn JavaScript on to hover a day, pick a window and draw your line. The numbers above come from the last build, Oct-26 2026 10:55Z.
What the tape says today · what's new · all 111 metrics