At 14:42:24 UTC on 23 August 2026, the Polymarket four-hour up/down market for HYPE printed below 0.20 for the first time. Over the next 26 seconds, ETH, BNB, DOGE and SOL did the same. Bitcoin's market took another 37 minutes.
We do not know what caused it, and this is not an attempt to guess. What the data shows is something narrower and more useful: how quickly a set of independent prediction markets agreed, and what happened to the liquidity underneath while they did.
Twenty-six seconds across five markets
All eight coins in this contract window opened at exactly 0.500 — four-hour up/down markets start at even odds by construction. Here is when each first printed below 0.20:
market first print below 0.20 lag
HYPE 14:42:24 —
ETH 14:42:41 +17s
BNB 14:42:48 +24s
DOGE 14:42:50 +26s
SOL 14:42:50 +26s
BTC 15:19:23 +36m 59sFive of six inside half a minute. These are separate order books with separate participants; nothing mechanically links them. The clustering is traders acting on the same information at roughly the same moment.
Bitcoin is the outlier and it is the interesting one. Its market did not cross 0.20 until 15:19:23 — 37 minutes later. Whatever moved the others either did not apply to BTC or was not believed to.
What the order book did
This is the ETH Up token through that single minute — best bid, best ask, and total size resting on each side:
time bid ask spread bid depth ask depth
14:42:00 0.750 0.860 0.110 32,744 5,142
14:42:10 0.730 0.840 0.110 32,599 5,317
14:42:20 0.730 0.810 0.080 32,178 4,941
14:42:23 0.690 0.710 0.020 27,123 5,733
14:42:26 0.530 0.690 0.160 19,976 15,813
14:42:30 0.500 0.520 0.020 18,309 23,672
14:42:36 0.200 0.210 0.010 7,780 24,858
14:42:41 0.140 0.170 0.030 3,480 28,922
14:42:50 0.050 0.080 0.030 2,657 26,844
14:42:59 0.110 0.150 0.040 1,171 30,390The price fell from 0.750 to 0.110. Three things happened underneath it.
- The bid side emptied. 32,744 shares resting at 14:42:00, 1,171 by 14:42:59 — a 96% fall. Buyers did not lower their prices so much as leave.
- The ask side filled. 5,142 shares to 30,390, up almost six times. Sellers arrived in volume, and they arrived within about ten seconds of each other around 14:42:26.
- The book inverted. At the start, resting buy interest outweighed sell interest roughly six to one. A minute later sellers outweighed buyers twenty-six to one.
The spread is worth watching too. It went 0.110, then 0.200, then 0.020, then 0.210, then 0.010 — repeatedly blowing out and snapping shut as the two sides tried to find each other. A one-minute average of that minute would read as an orderly move at a moderate spread. It was not orderly.
Then they disagreed
Everything fell together at 14:42. What happened afterwards split cleanly:
market opened closed
ETH 0.500 0.988
SOL 0.500 0.995
XRP 0.500 0.990
DOGE 0.500 0.600
BNB 0.500 0.075
HYPE 0.500 0.006
BTC 0.500 0.001ETH, SOL and XRP recovered from below 0.20 to near-certainty in the other direction. BTC, HYPE and BNB did not. DOGE ended almost where it started.
So the synchronised drop was not a market-wide verdict on crypto. It was a moment of shared uncertainty that resolved differently per asset over the following hour — with the assets that fell hardest at 14:42 among those that recovered furthest.
Why this needs second-level depth
Two of the three findings above are invisible at coarser resolution.
- The 26-second spread across markets cannot be measured in one-minute buckets, because every one of those five prints lands in the same bucket. At minute resolution this is one event with no internal ordering, and the question of which market moved first has no answer.
- The depth inversion needs the resting book, not the traded price. A trade log shows what filled; it does not show 30,000 shares of sell interest arriving, because most of it never filled.
The price path itself is the one part you could reconstruct from Polymarket's own free API, which serves 1-minute price history and is perfectly good for that — we say so plainly. The 37-minute BTC lag would show up there. The 17-second HYPE lead would not.
A later incident pushed the same two findings further. On 10 September, three "Bitcoin above $X at 12PM ET" strikes all printed 0.500 in the same second — a mid that was never executable, because the book behind it had stopped quoting. The strike ladder post works through the mechanism.
What we are not claiming
We have not identified a news event, and we deliberately have not gone looking for one to attach. Fitting a cause to a price move after the fact is easy and almost always wrong; the interesting measurement here is the timing and the liquidity, both of which stand without knowing why.
This is also one window on one day. The synchronisation may be typical or may be unusual — one observation cannot distinguish those, and we have not measured enough windows to say.
Checking it
Every figure here comes from four-hour up/down contracts ending 23 August 2026 16:00 UTC, readable through the order book history endpoint at 1-second resolution. The contract slugs are eth-updown-4h-1787486400 and the same pattern for each coin.
A published sample of 897,192 snapshots across 805 resolved markets, including four-hour contracts, is on Zenodo under CC BY 4.0 with no signup — listed here with the other downloads. The specific window above is outside that sample, so reproducing this exact minute needs the API rather than the file.