92¢ to Zero in 25 Minutes
FURIA vs FUT Esports, CS2, Esports World Cup playoffs, 22 August 2026
At 16:44 UTC, the market gave FURIA a 92% chance to win their Esports World Cup series. At 17:10, nobody would pay a single cent for their shares.
This is Polymarket's order book for that series. The desk holds 548,830 full-depth snapshots of this one book from match day, and each ridge is one of them, 3 minutes apart. Blue is money waiting to buy FURIA, orange is money waiting to sell, and the white dots are the price.
What the book did
At the scheduled start, 92% of the money near the price disappeared in 30 seconds. About $617k was at stake within 10¢ of the price at 13:59:59, and $48k at 14:00:29.
The book sped up about 67 times. Before the game it changed about 100 times a minute. In the last 30 minutes it changed about 6,700 times a minute, roughly 110 times a second, on a contract that can only trade between 0 and 100¢.
The deciding map swung the price across the whole board. FUT took map 1, FURIA took map 2, and late in map 3 the price went 92¢ → 13¢ → 67¢ → 1¢ → 16¢ → no bids, all within 25 minutes.
The question behind it
My research question is what an order resting in a book like this earns or loses while the price swings like that. A forecast is scored against the result. A resting order is scored against the spread, the fee, its place in the queue and everything that happens while it waits, and a book like this one shows how much can happen in that time. The research desk is built around that question.
How it is drawn
I drew the chart with matplotlib from the raw snapshots. Each ridge shows the money at stake at each price within 25¢ of the price, lightly smoothed over neighbouring cents and compressed (money^0.35) so the thin in-play book stays visible. A buy at 60¢ stakes 60¢ a share, and a sell at 60¢ stakes 40¢. There is one ridge every 3 minutes from 13:32 to 17:08 UTC, and the last bid was gone at 17:09:55.
The desk behind it
The research desk runs on a codebase I started on 16 August 2026, and AI coding agents wrote most of it, under rules that run as tests (how that works). As of 3 October 2026 it has 1,350+ commits, 280+ experiments and a 657 GB data lake. Its replay engine runs 344 strategy versions over 3,023 esports matches and gives byte-identical results when you run it twice.
Related: The research desk · How I run AI coding agents on one repo · The gate that keeps capital safe