The tape: orders, merging, pace

The tape (time & sales) is the stream of already executed trades in time order: time, price, size and the aggressor's side. One tape row is one print, the exchange's record that so many coins changed hands at such a price.

The side on the tape is the side of whoever came with a market order. “Buy” means the taker bought, taking a sell limit order from the book. “Sell” means the taker sold into a buy limit order. What was bought in a trade always equals what was sold: every buy on the tape has a seller, who was simply resting in the book and waiting. The tape tells you who was in a hurry, not who was more numerous.

The main difference between the tape and the book of sections 3–4: the book is intentions, the tape is facts. A wall can be pulled; a print cannot be undone. If a $5M buy went through on the tape, that money really was spent.

Why one order arrives as a dozen trades. The exchange's mechanism — the matching engine — fills a market order against the book's limit orders in turn: best price first, and within a price, whoever queued first. Each pair “taker — one limit order” is a separate trade, and the exchange sends it as a separate print.

Illustrative example. A 4 BTC market buy enters a book with sells of 0.5 BTC at 60,000, 1.2 BTC at 60,005, 2.0 BTC at 60,010 and 0.3 BTC at 60,020. That is four levels — and at least four prints. And if at 60,010 there are not one but thirty orders from different people, there will be thirty prints at that level alone. A $3M market order that went through ten levels may arrive as dozens of trades of $50–300K, and none of them looks large on its own.

A print's time comes in two kinds. Exchange time is the moment the engine executed the trade; all prints of one market order get the same time to the millisecond because they were executed by one action. Receive time is when the message reached us; it depends on the network, queues and the venue, and messages from different exchanges arrive interleaved. To answer “is this one order or several”, only exchange time is any good.

One more feature: each exchange has its own tape. Binance, Bybit and OKX trade the same coin at the same time, but they are different books and different engines. A buy on one venue and a buy in the same millisecond on another are two different orders, even if one person sent both.

To see orders rather than fragments, EVA merges trades back. There is one rule: trades of one exchange, one coin and one side with the same exchange time to the millisecond are one order. All four conditions are required. A buy and a sell in the same millisecond are different orders (they have different takers). Buys on Bybit and on Binance in the same millisecond are different orders (different engines). Two buys on Bybit a millisecond apart are different too.

In EVA's tape a merged row carries the Σ sign with a number before the price: “Σ7” means the order was assembled from seven trades. Above the tape it says “merged by time and side”, and the row's tooltip shows the price range the order went through. The price in the row is the average fill price, the amount is everything the order took from the book.

The average price of a merged order is weighted, not a plain average of prices. The illustrative example from the figure: 0.5 BTC at 60,000, 1.2 BTC at 60,005, 2.0 BTC at 60,010 and 0.3 BTC at 60,020. Dollar amount: 30,000 + 72,006 + 120,020 + 18,006 = 240,032. Quantity: 0.5 + 1.2 + 2.0 + 0.3 = 4 BTC. Average price — 240,032 / 4 = 60,008. The plain average of the four prices would give 60,008.75 — close, but wrong: only 0.3 BTC were bought at 60,020, so it must weigh less. On the tape this order is one row: Σ4, 60,008, 4 BTC, about $240K.

Merging is an approximation with two known limits. First: two different orders of one side on one exchange in the same millisecond cannot be told apart and become one row. In a calm market this is rare; in a sharp move it happens. Second: a large participant who does not want to be seen slices the order themselves — a piece every second or every minute. Such pieces have different exchange times, and merging will not join them: they honestly are different orders, just from one owner. We do not disclose merge windows and thresholds; what matters is the rule, and it is fully described above.

And a historical note: records from before merging was rolled out are per individual trade. The number of large rows “before” and “after” cannot be compared directly.

The right panel of the “Book · Flow” tab has the “Large trades” block — a tape of already executed large market orders: time, coin, side, price, amount and venue. “Large” here means rare in size for its venue and coin. The threshold is specific to each “exchange — coin” pair and it moves: it follows what is normal for that venue right now.

So a $300K order on a quiet venue makes it into the tape, while a $900K bitcoin order on Binance may not: that size is ordinary there. Next to each row it says which threshold it passed and how many times over — the multiple. Illustrative example: venue threshold $200K, order $500K — multiple 500 / 200 = 2.5. On another venue the threshold is $2M, the order $900K — it does not make the tape, although in dollars it is almost twice the first one.

The multiple is the main number of the row: it is comparable across venues and coins. A ×8 order on a quiet venue is more unusual for its market than ×1.2 on the most liquid one.

Do not confuse this tape with the “Large trades” layer on the chart. The layer's threshold is chosen in the “▾” menu — $500k, $1M, $2M, $5M, $10M — and it is a “show from” filter, the same for all venues. The panel tape's threshold is specific to each venue and moves. So the same order can be a dot on the chart and not a row in the tape, and vice versa. That is not a bug but two different questions: “where did big money go through” and “what is unusual for this venue”.

What else the tape does not show. Who was behind an order — we do not know: the public stream has no names. Whether it was one person or a bot slicing an order — we do not know. What they will do next — even less. And every trade has a second side: if a taker bought $5M, someone sold them the same $5M with a limit order. The tape shows only the one who was in a hurry.

A series of large buys at one price where the chart shows a sell dense order is no longer an intention but an execution: the wall is being eaten. This is where the tape and the book work together: the book says what rests, the tape what happens to it.

Above the large-trade tape EVA shows the tape pace: how many trades per second are happening now and how many times that exceeds the coin's usual. The line reads in words: “tape is boiling”, “picked up”, “quieter than usual”, “almost stopped”. Under the chart there is the “Tape activity” band with the same ratio: ×1 — as always, ×3 — three times denser.

Why a ratio and not a raw number. “Six trades a second” means nothing by itself: for bitcoin that is a lull, for a small coin a storm. The ratio to the coin's own norm is comparable across coins and with its own yesterday. Illustrative example: a coin's usual pace is 15 trades a second, now 45: 45 / 15 = ×3. Another coin's usual pace is 2 a second, now 6 — also ×3, though with seven and a half times fewer trades.

Pace is counted by number of trades, not by money. This matters: a hundred small trades a second is panic or bots, even if little money is in them; one large order is calm position building, even if it holds $10M. So pace and large orders are different readings, and you look at them together. The ratio is computed over the last closed minute: an unfinished minute would understate the pace every time for no reason. And in the first minutes after a start there is nothing to compare with — the ratio is not computed then.

Useful combinations. Pace is up, large orders go one way, and the price stands still — someone is meeting that flow with limit orders: the chart often shows a dense order there that is being eaten and topped up (section 4). How it ends, pace will not say, but the place where it happens is worth remembering. Pace is up but there are no large orders — small participants or bots are moving it. Pace dropped almost to zero in the middle of the day — the market is waiting for an event, and the first large print in such quiet moves the price more than usual: the book is thinner in a lull.