PredictionManager

The PredictionManager provides states, callbacks, and settings to fine tuning prediction for your game type.

Description

The PredictionManager is a core component responsible for managing the prediction system, which allows clients to predict their own actions and then reconcile (correct) their state when authoritative updates are received from the server. It allows you to tweak various global prediction settings and get certain prediction callbacks.

Check out its API page for more specific methods and events here.

Settings

Default Settings

⚙️ Create Local States

When enabled this allows the client to create local reconcile states which means reconciles can be sent less frequently and provide data to use for reconciles when packets are lost.

⚙️ State Interpolation

This is how many states to try and hold in a buffer before running them on clients. Larger values add resilience against network issues at the cost of running states later.

⚙️ State Order

The order in which clients run states. Future favors performance and does not depend upon reconciles, while Past favors accuracy but clients must reconcile every tick.

⚙️ Drop Excessive Replicates

Toggle the setting to true to drop excessive client replicates, helping prevent attacks but risking temporary desynchronization during connectivity issues. Set to false to store up to 3 seconds of replicates, processing multiple per tick for quicker cache clearing. This approach ensures all inputs are executed but could enable speed hacking. When false, replicated data is consumed more quickly if the cache limit is exceeded.

⚙️ Maximum Server Replicates

This is shown when the Drop Excessive Replicates is enabled. This value indicates the maximum number of replicates which the server will cache from client before dropping old ones. Higher values will reduce the chance of dropped input when the client's connection is unstable, but will potentially add latency to the client's object both on the server and client. Generally, cached replicates will never significantly exceed Queued Inputs.