Two masked figures on a wet rooftop, a city of signage behind them, one lantern burning. Neither of them can be hit until they commit, and committing takes two hundred and forty milliseconds. The parry window is fifty. Everything about whether this game works or does not work is decided by arithmetic before either player touches a button.
Small match, large budget per player
There is an obvious economic argument for high tick rates in duelling games that almost nobody acts on: with two players instead of four hundred, the server cost of resolving the world frequently is trivially small. Rooftop Duel is one of the few titles that has followed the reasoning through.
A hundred and twenty ticks gives an interval of eight and a third milliseconds. With our near-region round-trip time of thirty-two, the peek advantage lands at forty-one — inside the fifty-millisecond parry window, which means a parry that looks correct on your screen is a parry the server will usually agree with.
| Connection profile | Median RTT | Peek advantage | Disputed parries |
|---|---|---|---|
| Fibre, near region | 32 ms | 41 ms | 3% |
| Cable, near region | 49 ms | 57 ms | 6% |
| Congested shared line | 91 ms | 99 ms | 14% |
| Throttled tether | 184 ms | 192 ms | 31% |
The indicator is the best in the archive
When the server overrules your client, a thin bar flashes at the base of the screen and the replay marks the frame. It is unmissable without being intrusive, and after about twenty matches our testers stopped noticing it consciously while still, demonstrably, adjusting to it.
That adjustment is the point. On the throttled tether, where thirty-one per cent of parries go against the local view, testers who could see the indicator described the game as hard rather than broken. In Verge Squad, with corrections at a quarter of that rate and no indicator at all, the same testers used the word cheating.
Two games, the same underlying event, opposite verdicts. The difference was a bar at the bottom of the screen.
Four regions and a queue that shows its working
Four hosting locations, and a matchmaking screen that displays your measured round-trip time to each before you commit, alongside an explicit warning when nothing under eighty milliseconds is available. You can accept a worse match; you cannot accidentally take one.
Only Slipstream Nine does this as well, and between them they represent the entire set of games in our archive that treat the player as capable of understanding a number.
Where it loses its marks
The population is thin outside two of the four regions, and the honest queue makes that visible in a way that is discouraging even though it is correct. Waiting four minutes to be told there is nobody within eighty milliseconds is better than being thrown into a bad duel, but it is still four minutes.
And there is no reconnection window at all. A dropped connection is a forfeit, full stop, on a game where matches run six to nine minutes. Spore Depths holds a slot for four minutes; even forty-five seconds here would have saved us nine forfeits on the throttled profile.
| Measure | Rooftop Duel | Archive median |
|---|---|---|
| Tick rate | 120 Hz | 28 Hz |
| Hosting regions | 4 | 4 |
| Visible corrections per match | 0.7 | 3.6 |
| Correction indicator | Yes | None |
| Reconnection window | None | 45 s |
The short version
- A hundred and twenty ticks for a two-player match, which is the correct extravagance.
- Peek advantage of forty-one milliseconds against a fifty-millisecond parry window.
- The clearest correction indicator we have tested, and the effect on perception is measurable.
- Latency to all four regions shown before you queue.
- Thin population outside two regions, made visible by the honest queue.
- No reconnection window whatsoever, which is an unforced error.
Verdict
Eight point six. A small game that spent its entire technical budget on the two things that decide whether duelling works online, and then had the confidence to tell players the truth about both.