Rollback Netcode: Why Fighting Games Use It

Rollback netcode is the gold standard for online fighting games, making matches feel responsive and fair despite latency where the older delay-based approach felt laggy and inconsistent. Understanding why rollback works so much better for fighting games explains why it's become the expected standard.

Rollback predicts and corrects instead of waiting

Fighting games demand precise, responsive inputs, which makes latency especially damaging. The old approach, delay-based netcode, added input delay to wait for the opponent's inputs, which made the game feel laggy and inconsistent as latency varied. Rollback netcode takes a different approach: it predicts the opponent's inputs (usually assuming they keep doing what they were), simulates forward immediately for responsiveness, and when the opponent's real inputs arrive, rolls back to that point and re-simulates with the correct inputs if the prediction was wrong. This means the game runs responsively (no input delay, immediate simulation based on prediction) and stays correct (rollback corrects mispredictions), so the local player gets responsive, consistent input regardless of latency, with the netcode hiding the latency through prediction and correcting through rollback. Predicting and rolling back, rather than adding delay and waiting, is what makes rollback feel responsive where delay-based netcode felt laggy.

Rollback's corrections are usually invisible, which is why it's so good for fighting games. The reason rollback works so well for fighting games is that its corrections are usually small and visually minor: predictions are often correct (players frequently continue their inputs), and when wrong, the rollback re-simulates only a few frames, producing a small visual correction that's usually imperceptible in the fast action of a fighting game. So the local player experiences responsive, consistent input (the key requirement for fighting games), and the occasional corrections are minor and rarely noticeable, giving a far better experience than delay-based netcode's constant input lag. This is why rollback has become the expected standard for online fighting games: it provides the responsive, consistent input the genre demands, hiding latency through prediction and handling it with mostly-invisible rollbacks, where delay-based netcode imposed the laggy, inconsistent input that made online fighting games feel bad. Rollback is more complex to implement (requiring the game to be able to save state, roll back, and re-simulate deterministically, which connects to the importance of determinism), but the responsive, fair experience it provides is so much better for fighting games that it's become the standard players expect. Understanding why rollback works—predicting and rolling back for responsiveness with mostly-invisible corrections—explains why it's the gold standard for online fighting games and why its complexity is worth it for the genre.

Consistency beats intensity

Indie development is a long game, and it rewards steady, sustainable effort more than heroic bursts. A little progress made consistently — on the game, on the marketing, on the community — compounds in a way that last-minute sprints never do. The developers who finish and find an audience are usually the ones who kept showing up, not the ones who worked themselves into the ground for a week and then burned out.

Build a pace you can sustain, and protect it. Momentum is fragile and expensive to rebuild, so steady forward motion is worth more than any single intense push.

Let real players be the judge

It's remarkable how differently real players behave from how you imagine they will. The tutorial you think is obvious confuses them; the feature you agonised over goes unnoticed; the thing you almost cut becomes their favourite. None of that is visible from inside your own head, which is why watching real people play is the single highest-leverage thing most developers under-do.

Watch without intervening, resist the urge to explain, and pay attention to what players do as much as what they say. Their confusion and their choices are data, and acting on that data is what turns a game that works for you into one that works for everyone.

Polish where players actually look

Polish is not evenly valuable. Players form an impression in the first minutes and spend most of their time in the core loop, so effort spent there returns far more than effort spread thin across content few people reach. The opening, the moment-to-moment feel, and the things every player touches are where polish converts directly into how good the game feels.

Be deliberate about it. Make the first impression strong and the core interactions satisfying before widening out, because a great core with less content almost always beats a sprawling game that never feels good to play.

Scope is a decision, not an accident

Almost every overscoped game got that way one reasonable addition at a time, with no single decision ever feeling like the mistake. The finish line recedes a little with each new feature, and because the project always feels nearly done, the developer rarely notices how far the goal has drifted until they're exhausted and the game still isn't out.

Treat scope as something you actively decide rather than something that happens to you. Write down what the finished game contains, make every addition a conscious trade against that, and keep most new ideas in a backlog where they belong — because a small game you finish beats a large one you abandon.

Measure before you optimise

Intuition about what's slow, what's confusing, or what's driving players away is usually wrong, and acting on it wastes effort on problems that don't matter while the real ones persist. The developers who improve their games efficiently are the ones who measure first — profiling performance, watching real sessions, capturing actual errors — and let the data set their priorities.

It's slower than trusting your gut, but it's the only approach that reliably improves the game instead of just changing it. Find the biggest real problem, fix that, and measure again, rather than optimising guesses.

Rollback netcode predicts the opponent's inputs and rolls back to correct when real inputs arrive, giving responsive, consistent input despite latency with usually-invisible corrections. It's the gold standard for online fighting games, far better than delay-based netcode.