If you have ever watched a Bitcoin order book heatmap, you know the feeling: a huge bid appears a little below price, it looks like a floor, and then, right before price gets there, it is gone.
We wanted to know how often that really happens, so we measured it. This post explains how we classified every large wall in the Binance BTCUSDT perpetual order book as held, eaten or pulled, what the numbers were, and what they do not prove.
Disclosure: I work on Swiftedge, a free browser tool that shows the live Bitcoin order book as a heatmap. The data below comes from our own recordings. The full write-up is on our site.
The data
- Our own recording of the Binance BTCUSDT perpetual order book, four snapshots per second, plus every trade.
- Period: 22 September to 18 October 2026 (about four weeks).
Step 1: what counts as a "large wall"
A wall is a 25 USD price level whose resting size is in the top 3 % of that day, and which stands for at least six minutes. The six-minute rule removes the flicker of orders that appear and vanish within seconds.
A wall disappears at the moment its size falls below 30 % of its earlier size and stays there.
Step 2: held, eaten or pulled
The key idea is to look only at the trades at the wall's own price. If size vanished from the book, either someone traded against it or someone cancelled it. The trade tape tells you which.
- Eaten: at least half of the size that disappeared actually traded.
- Pulled: at most 10 % traded, the wall was at least 3 USD from price, and at most 10 % traded over its whole life.
- Held: price reached the wall, at least 5 BTC traded into it, most of it was still there afterwards, and price then turned at least 60 USD away.
- Mixed: everything in between. Slow fades and walls with incomplete data were left out.
801 large walls ended while price was within 0.4 %. 244 of them could be classified cleanly, and the results below describe those 244.
The results
| Outcome | Walls | Share | Median size | Median distance from price | Median time to disappear |
|---|---|---|---|---|---|
| Pulled | 116 | 48 % | 123 BTC | 151 USD | 0.3 s |
| Eaten | 99 | 41 % | 127 BTC | 3 USD | 0.4 s |
| Mixed | 29 | 12 % |
A few things stood out:
- Pulled walls go fast. 71 % vanished within one second, and the median one lost 97 % of its size in that moment.
- Pulled walls rarely see a trade. In at least nine of ten cases, not a single BTC traded at the wall's price during its life.
- They were not fleeting: a pulled wall had typically stood for 13 minutes (middle half: 7 to 23 minutes) before it was cancelled.
- Bids and asks were pulled about equally often (55 bids, 61 asks), and size did not change much: 41 % to 58 % of walls were pulled in every size group.
Distance is the strongest clue
For the 186 walls that were still in front of price when they disappeared:
| Distance from price | Walls | Pulled | Eaten |
|---|---|---|---|
| Under 25 USD | 61 | 20 % | 75 % |
| 25 to 75 USD | 37 | 62 % | 27 % |
| 75 to 150 USD | 25 | 88 % | 0 % |
| Over 150 USD | 63 | 94 % | 3 % |
Of walls more than 75 USD from price, 92 % were pulled. Close to price, walls mostly get traded into. Far from price, they mostly get cancelled before price ever arrives.
And real "holds" were rare: of 172 walls that price came within 37.5 USD of, only 9 met every hold condition.
One example
On 18 October 2026, a bid wall of about 416 BTC (roughly 34 million dollars) sat at 81,960, about 34 USD below price, for 30 minutes. At 11:54:02 UTC it was cancelled in 0.3 seconds. Over its whole life, only 7 BTC had traded at that price.
What this does not prove
- A cancelled order is not proof of spoofing. Market makers and large traders cancel and re-place orders all the time, for many reasons. The order book shows what happened, not why.
- Only 244 of 801 walls could be classified cleanly.
- It is one exchange's perpetual market over four weeks.
- It is descriptive and looks backwards. Nobody can tell in advance whether a given wall is real.
See it yourself
You can watch walls appear, get eaten and get pulled in real time in Swiftedge Depth: the live Bitcoin order book as a heatmap, every trade as a bubble, plus options GEX levels. It is free and runs in the browser without an account.
Market data and descriptive analysis only. Not financial advice.
Top comments (1)
The distance split is useful. I'd add the exclusion count beside each distance bucket. Only 244 of 801 walls entered the outcome table, so the 557 excluded cases could matter if missing data is concentrated at particular distances. I'd keep a wall whose record ends before its outcome as unresolved, rather than treating its disappearance as a pull. Were incomplete recordings more common near the touch or farther away?