Client-side prediction: move before the reply

A responsive player and an authoritative server do not have to wait for each other on every frame.

The round trip is a waiting room

Without prediction, a client sends an input and waits until a server snapshot shows its effect. The wait includes the outward trip, server scheduling and the return trip. A fast connection helps, but a smooth-looking animation alone cannot remove this causal delay.

Predict the input you already know

The client knows the movement you just requested. It can apply that movement locally immediately, while sending the command to the server. That is prediction of your own input, not prediction of another player’s future decision. The server remains responsible for the authoritative result.

Try a controlled comparison

Keep latency at 180 ms, jitter at 30 ms and packet loss at 2%. Play the example once with Prediction on, then turn it off. Scrub around the first dodge input. Compare the solid local circle with the dashed authoritative circle. The input sequence stays the same; you are changing how the client presents it.

What prediction cannot promise

Prediction does not make a packet arrive sooner. A collision, server rule or disagreement in movement can still cause a correction. In this small model the movement rules are identical and there are no obstacles that block the player, so prediction-only can look accurate more often than it would in a complex game.

Read the result carefully

An immediate local dodge and a server-confirmed hit can coexist. One is a presentation of local inputs; the other is a decision based on the commands the server received. Lag Arena is a teaching model, not a diagnosis of why you lost a particular match.