RizonFixed windows vs horizon contracts: why arriving late costs you
Arrive at a random moment in a 5-minute Bitcoin window and the side ahead is already worth 74¢ (median). Horizon contracts start at your fill instead. The data.
Short answer: In a fixed window, everyone shares one start and one end, so the later you arrive, the less time you get and the more the price has already moved from the price to beat. Measured over every second of 8,633 five-minute Bitcoin windows (Sep–Oct 2026), a random arrival found half the window gone, the price a median $23 from the open, and the side that was ahead already worth a median 74¢; in 57% of arrivals one side was worth 70¢ or more. A horizon contract starts at your fill and runs its full length, at a strike next to the price, so every arrival gets the same contract.
Two ways to sell "up or down"
Fixed-window markets cut time into blocks on the clock: Bitcoin up or down from 3:15 to 3:20, then 3:20 to 3:25, and so on. Each block has one opening price, the price to beat, and one close. Polymarket lists Bitcoin windows of 5 minutes, 15 minutes and 4 hours (each "Bitcoin Up or Down - October 27, 8:50PM-8:55PM ET", settling on a 60-second Chainlink average at each end; rules), and Kalshi lists 15-minute ones (rules). A horizon contract has no blocks: it starts when you trade and runs for a set time (15 seconds, 1 minute or 5 minutes on Rizon), against a strike next to the price at your fill.
| Fixed window | Horizon contract | |
|---|---|---|
| Starts | At the window's open, the same for everyone | At your fill |
| Time you get | Whatever is left of the window | The full horizon |
| Strike | The price at the window's open | The book's strike, next to the price when you trade |
| Price | Starts near 50¢ and drifts toward 0¢ or 100¢ as the window ages | Fixed: pay your size, get 2× back if you're right |
| Arriving late means | Less time, a strike set before you came, a price that has moved | The same contract as anyone else's |
| Expiry | Shared by the whole window | Your own, a fixed time after your fill |
What a late arrival actually finds
We treated every second of 30 days of Bitcoin data (2.59 million seconds, Binance BTCUSDT) as an arrival in its clock-aligned 5-minute window, and measured three things at that moment.
Time left. Arrivals spread evenly through the window, so the median arrival has 2 minutes 30 seconds left, 20% have under a minute, and 10% have under 30 seconds. That's arithmetic, not data, and it's easy to skip.
How far the price has already moved. The median arrival found Bitcoin $22.92 from the window's opening price; one in four found it more than $49.74 away, and one in ten more than $88.18. In units of a whole window's typical move (one sigma over 5 minutes), the median was 0.41 and the 90th percentile 1.42.
What the side that's ahead is worth. Once the price has moved, the side it moved toward is more likely to win, and its fair price rises. Measured from the data (how often windows in the same situation finished that way):
| Fair price of the side ahead (¢) | Share of arrivals (%) |
|---|---|
| 50–60 | 24.6 |
| 60–70 | 18.5 |
| 70–80 | 18.3 |
| 80–90 | 15.2 |
| 90–100 | 23.4 |
The median arrival met a leading side worth 74¢. In 56.9% of arrivals it was worth 70¢ or more, in 38.6% 80¢ or more, and in 23.4% 90¢ or more. Only a quarter of arrivals found something close to a coin flip.
The later you arrive, the less is left to trade
| Time into the window (m:ss) | One side worth 70¢ or more (% of arrivals) | One side worth 90¢ or more (% of arrivals) |
|---|---|---|
| 0:00–0:10 | 2.1 | 0.0 |
| 1:00–1:10 | 36.5 | 3.0 |
| 2:00–2:10 | 56.4 | 11.2 |
| 2:30–2:40 | 62.8 | 17.6 |
| 3:00–3:10 | 69.1 | 24.7 |
| 4:00–4:10 | 80.3 | 45.3 |
| 4:30–4:40 | 87.7 | 62.6 |
| 4:50–5:00 | 93.9 | 82.8 |
Arrive in the first ten seconds and the window is still a fair question. Arrive at 2:30 and in nearly two cases out of three one side already costs 70¢ or more. In the last ten seconds, 83% of the time one side is priced at 90¢ or more: you're mostly paying for an answer that's already in.
None of this makes a late trade wrong. A fair 80¢ price is fair. But it's a different trade from the one the window advertised: less time, a lopsided payout (pay 80¢ to make 20¢), and a strike chosen by the clock rather than by you.
Fair prices have fat tails
A normal random walk would say a side 2 to 2.5 sigmas ahead (of the time left) wins 98.6% of the time. In the data it won 94.6%. Leads of 1.5 to 2 sigmas held 91.8% of the time against a normal model's 95.7%. Bitcoin's short-horizon moves have more large jumps than a bell curve, so heavily favoured sides lose more often than the textbook says. If you price a late window with a normal model, you'll overpay for the favourite.
How a horizon contract changes the arrival
- Every arrival gets the full time. A 5-minute contract bought at 3:18 runs to about 3:23, not 3:20.
- The strike starts next to the price. You choose up or down at a strike near spot, not at a strike set minutes before you arrived. That's why every Rizon contract can cost the same: pay your size, get 2× back if you're right, the equivalent of a 50¢ contract.
- The edge is in the strike. Instead of paying 74¢ for the side that's ahead, you decide whether the next five minutes beat a level that's about a coin flip, and the book competes on how far that level sits from spot (see strike vs probability).
- There's no common close to race. Each contract expires on its own second, so there's no single final moment where everyone's position is decided at once (see how often the last 30 seconds reverse a window).
How we measured it
- Data: Binance BTCUSDT 1-second klines (archive), 27 Sep to 26 Oct 2026. Windows aligned to the clock; the opening and final prices are the last trades at or before each boundary.
- Single prices: each window's opening and final prices are single last trades. Venues that settle on 60-second averages set the price to beat from the minute before the open; the arithmetic of a late arrival (less time, a price that has moved) is the same.
- Fair price: for each arrival, the lead over the open measured in sigmas of the time left (sigma from 10-second returns over the 30 minutes before the window opened), then the share of all arrivals with a similar lead whose window finished on that side. A window that finished exactly at its open counts half. This is a historical frequency, before spreads and fees, and it's what a well-calibrated market would charge on average.
- Rerun:
bun studio/research/late-arrival.tsin Rizon's repository.
FAQ
What is a fixed-window up/down market?
A market on whether an asset finishes a block of clock time (say 3:15 to 3:20) above or below its price at the block's open. Everyone in the block shares the same start, the same price to beat and the same end.
What is a horizon contract?
A contract that starts when you trade and expires a fixed time later: 15 seconds, 1 minute or 5 minutes on Rizon's Bitcoin markets. Its strike is set on the book at your fill, so it starts close to the price.
How much does arriving late in a 5-minute window cost?
Time and price. A random arrival has 2.5 minutes left on median, and the side ahead is already worth a median 74¢; past 2:30, one side is worth 70¢ or more in nearly two arrivals out of three.
Is buying the side that's ahead late in the window a good trade?
Only if it's priced below its real chance, the same as any trade. Our data suggests big favourites lose more often than a normal model predicts (94.6% wins where a normal model says 98.6%), so heavily favoured sides are easy to overpay for.
Why does Rizon use horizon contracts instead of windows?
So that time starts when you do: every contract runs its full length from your fill, at a strike next to the price, at the same 2× payout, whenever you arrive.