Dollars of bitcoin changing hands each day, venue by venue, spot and perpetuals kept apart — from the exchanges' own daily candles, back to 2013. One principal book per venue; steps in a total are venues joining, not the market moving.
$6.33B spot traded each day, venue by venue and added up · every venue added up on Oct-22 2026, the last day in this build, read from the venues' records, published by askthetape.com.
Every exchange publishes a daily candle for its own bitcoin book, and inside that candle is the money that changed hands. Nobody has to be asked for it and nobody has to be paid: it is the same public handle a chart on their own site reads, and it goes back years — Bitstamp spot to January 2013, Bitfinex spot to March 2013, Coinbase spot to July 2015. So this page is not our capture spliced onto somebody's history; it IS the history, 23 venues of it. On 2026-10-22, 14 spot books traded $6.33B of bitcoin between them and 13 perpetual books traded $47.85B — 7.6 dollars of perpetual for every dollar of spot. The busiest spot book was Binance with 25% of the spot side; the busiest perpetual was Binance USD-M with 30%. Two warnings before you read a share. One: this is what the venues REPORT. Nothing here is verified against a chain or an auditor, wash trading is not filtered, and a venue that inflates its print inflates its share. Two: a «day» is not the same 24 hours everywhere. Every book's day was MEASURED on this pull, not read from documentation, and Bithumb's day opens 15:00Z · HTX's day opens 16:00Z — so 3 books on those clocks are drawn in the legend and left OUT of the totals above. Of the dollars that ARE in the totals, 9% of the spot side and 9% of the perpetual side come from books that publish only the coins traded, priced here at the day's typical price. Historical archive fetched 2026-10-06 from the venue's public endpoint / the public CFTC endpoint; source units and gaps preserved; browser-only.
| leg | each exchange's own public daily-candle handle, one principal bitcoin book per venue: 31 books across 23 venues. We show the UTC-day books in the totals and list books on other clocks separately, with both counts taken from the same captured list: 28 (14 spot, 14 perpetual) on a UTC day, 3 on other clocks. Pulled by `vx_fetch.py` at no more than two requests a second per venue. Spot is BTC/USDT, BTC/USD or BTC/KRW; perpetual is that venue's main BTC perpetual. Every endpoint, its unit and the date it was last probed sit in data/vx/volume_meta.json |
|---|---|
| 🩸 reported, not verified | reported volume is each venue's own print. It is not verified against anything, wash trading is not filtered, and the shares on this page are shares OF WHAT VENUES REPORT — not shares of real trading. A venue with an incentive to look busy can move its own share here and nothing on this page would contradict it |
| 🩸 one book per venue | one principal book per venue, and the rest of that venue's bitcoin books are NOT added. The size of what is left out is measured, not guessed: on 2026-10-22, Binance's dollar-quoted BTC books beside the one we carry (BTCFDUSD, BTCUSDC) added $361M against $1601M — 22.5% on top of the single book. Its local-currency books (BTC/TRY, BTC/JPY, BTC/EUR) print their turnover in those currencies and are not converted or added. Read every share as «share of one book per venue», not «share of all bitcoin trading» |
| units | 22 of the 31 books publish the dollar turnover themselves; the other 9 publish only the coins that changed hands, and those are priced at the day's typical price (high+low+close)/3. That approximation is measured, not assumed: run the SAME proxy on 5 books that publish BOTH and it lands within 0.21% of the venue's own dollar figure at the median (bias +0.008%, worst single day 3.3%) — measured 2026-10-23. 🩸 That 5-book calibration is run on the DEEPEST books on the page (Binance, Bybit, OKX, KuCoin), and the proxy is then applied to thinner ones — Hyperliquid, Coinbase Intl, Bitstamp, Bitfinex, Gemini, Crypto.com, Kraken Futures. On 2026-10-22 that is 8.8% of the spot total and 9.3% of the perpetual total priced by a proxy whose error was measured somewhere else; a thin book's day has fewer trades and can drift further from its typical price than a deep one. Korean books are quoted in won and converted at our own USD/KRW of the same day, the rate the Korea premium page prices with; contract books (Binance COIN-M) are converted at the exchange's own contract size |
| calibration | What is measured here is not a verification of the venue but the COMPLETENESS OF OUR OWN CAPTURE, and the page says so plainly: our minute-by-minute capture of the trade tape (cvd_1m on the production node) divided by the exchange's own daily candle, on the days both exist (2026-08-07 → 2026-10-22). Binance USD-M median ×1.0000 (min ×0.2624, p10 ×0.9726, p90 ×1.0000, max ×1.00000000) on 77 days, 35 of them exact to the cent, 1 above 1.0, 33 with holes in our own tape · Bybit median ×0.9993 (min ×0.2313, p10 ×0.9903, p90 ×1.0000, max ×1.00000000) on 77 days, 14 of them exact to the cent, 0 above 1.0, 63 with holes in our own tape · Gate median ×0.9987 (min ×0.2246, p10 ×0.9723, p90 ×1.0000, max ×1.00000000) on 77 days, 28 of them exact to the cent, 0 above 1.0, 46 with holes in our own tape · OKX median ×0.9983 (min ×0.2446, p10 ×0.9915, p90 ×1.0000, max ×1.00000000) on 77 days, 12 of them exact to the cent, 0 above 1.0, 65 with holes in our own tape. Read it the right way round: the ratio never once goes ABOVE one, and a large share of days land exactly on it — our tape is a strict SUBSET of the same venue print, one source read two ways, so this proves we lost nothing and proves nothing about what the venue published. Wash trading, if there is any, sits in both numbers. The days below one are holes in OUR capture, not in the venue's: the worst (2026-08-07) is 540 minutes of 1,440. We do not print a correlation for this pair — with one 540-minute hole in the window it reads as 0.95 and without it as 0.998–0.9999, which says more about the hole than about the agreement. Our capture is of the PERPETUAL tape, so the spot books on this page have no second reading behind them at all |
| 🩸 the unit guard | a field named «quote volume» does not always carry one — but a venue whose local price really is 20 % away from the world is not broken, it is Korea in January 2018. So the guard is written down in TWO stages before the answer is looked at, and both are applied to everyone. Stage one asks whether the book agrees with ITSELF: dollars ÷ coins against the (high+low+close)/3 of the SAME candle. Outside ±15 % the row is quarantined — that is a unit, not a market. Stage two runs only on rows that passed stage one: if the implied price is more than ±25 % from the median implied price of every other book that day, the row is KEPT and flagged a premium day. This build quarantined nothing, and KEPT 27 venue-days with a local price gap, the widest Upbit on 2018-01-08 at +59 %, Upbit on 2018-01-09 at +55 %. An earlier version of this page had only the cross-venue test, and it erased 26 Upbit days at the very top of the 2017 cycle — days on which Upbit was around a third of the whole spot side — while the page's own «how many venues that day» line printed the hole as a venue arriving. That was the wrong null hypothesis, not the wrong threshold: on 2018-01-08, the worst day by the cross-venue test, Upbit's own candle agreed with itself to 0.5 %. 🩸 The diagnosis stays a HYPOTHESIS: on Gate's 2015-2017 candles the quote field and base × price disagree by a factor of 1.46-1.63 that drifts smoothly to 1.0 by 2018 — that is not a currency (the won or yuan would be a constant ~6.2) and we do not know which of the two fields is wrong. Those years are excluded because the two fields contradict each other, not because anyone here has proven which one lies |
| composition | 🩸 a total is a sum over the books that EXIST that day, so a step in it is a venue arriving, not the market. The «how many venues that day» view carries that count as its own line, the chart marks the last steps, and beside the total runs a second line built from the 9 spot and 9 perpetual books that have run end to end without a gap — same venues on the first day and the last. Shares are printed only inside one composition stretch (2025-09-14 → 2026-10-22 for spot, 2026-10-01 → 2026-10-22 for perpetuals) and never across a seam |
| cadence | one row per book per day, taken from a candle that has already CLOSED. 🩸 The guard is on the candle's close, not on its label: a book whose day opens at 16:00Z has a bar labelled «yesterday» that is still running until 16:00Z today, and an earlier version of this page published exactly that — HTX's «Sep-13» read $261.9M at 13:11Z and $294.5M ninety minutes later, growing, because the bar had not closed. Nothing here is drawn until its own 24 hours are over, which is why the books on a foreign clock stop one day earlier than the UTC ones. For the 28 books whose day IS a UTC day, the page is one day behind by construction; for the other 3 it is one day behind plus the offset of their own clock. Pull of this build: 2026-10-23T21:23:50Z |
| 🩸 the daily top-up | the full history is pulled once; after that a top-up re-reads only the last three days and MERGES them into the file, one row per book per day. Two things it is not allowed to do, because on 2026-09-14 it did both for two hours: it may not drop a book (a handle that answers with nothing leaves that book's existing rows exactly as they were, and says so here), and it may not re-label days it did not fetch (the day convention measured for a row when it was first pulled stays with that row — three fresh bars do not get to re-date nine years of history). If a top-up would end with fewer books or fewer old rows than the file already had, the file is not rewritten at all and the age of the data shows up in the freshness check instead. Last top-up 2026-10-23: 30 of 31 books re-read, Coinbase Intl perp: last refreshed 2026-10-01 — the venue moved its perpetuals to Deribit on 2026-10-01, so this book ends there |
| 🩸 where the history stops | no venue is carried further back than its own handle goes, and the handles that refuse depth are named rather than filled: Bithumb spot — the handle takes no time argument: it returns the 200 most recent daily bars and nothing deeper; Gemini spot — the candles handle takes no time argument: it returns the 364 most recent daily bars and nothing deeper; HTX perp — the handle takes only `size` (max 2,000 bars): 2000 daily bars is everything it gives; HTX spot — the handle takes only `size` (max 2,000 bars), no time argument: 2000 daily bars is everything it gives; Kraken spot — the OHLC handle returns the last 721 daily bars only |
| 🩸 whose day is it | every book's day is MEASURED on the pull, never read from documentation: the writer records the opening stamp of every bar it parses and the meta carries the modal offset (`measured_open_utc`, with the number of bars it was measured on). 28 of the 31 books open at 00:00Z. We identify excluded books by their actual opening clock and keep the same exclusions in the text and totals: Bithumb spot opens 15:00Z · HTX perp opens 16:00Z · HTX spot opens 16:00Z. They are drawn in the legend and they are NOT in any total, share or venue count on this page. MEXC spot is in the totals from 2018-08-23; its 302 earlier days opened at 16:00Z and are drawn but not added. 🩸 Two books were fixed rather than excluded: Bitget publishes the same candle on a `1Dutc` handle (switched to it, bars now open 00:00:00Z), and Deribit's own 1D bar opens at 08:00Z, so its days here are built from the venue's HOURLY bars, 24 per day — on 2026-09-13 that is $140.7M against the $219.9M the 08:00Z bar would have handed us, a 56% difference that an earlier version of this page published as a UTC day |
| policy | public exchange endpoints only · derived aggregate by askthetape.com · no export · read on the site · not exported: this page is drawn from someone else's archive |
| 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-22 2026 |
| policy | no backfill · no export · read it here · read on the site · not exported: this page is drawn from someone else's archive |
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-23 2026 21:24Z.
What the tape says today · what's new · all 111 metrics