Sorry, I've written a bit of a novel here. They see close approximations, but there will always be points where their view will differ slightly from the server's authoritative copy, even with the best compensations in place. The most significant of these is that, due to the nature of predictive movement and interpolation used in fast-paced network games, not all users see the same presentation of the game world. There are obviously issues common to all network multiuser applications that will apply, but those are not unique to Net Rewinder. There are no Net Rewinder specific flaws/caveats I'm aware of beyond what's mentioned in the documentation. Since the local host won't be using rewinding, they'll be tested against live data rather than the recorded results, and so will have a slight advantage in this case. The one exception to this is if the server is also a client.

It should be noted, however, that hits from all clients will be tested against the same data, so they'll all get consistent hit/miss results. In the unlikely event that you need to reduce the recording fidelity for any reason (thousands of agents etc), rewind requests for timesteps between the snapshots will be interpolated from their neighbouring timesteps, and may diverge slightly from their actual historical values. This means that you can rewind any character to the exact historical position and orientation they had at any timestep in their history. If you use the default settings (appropriate for most use cases, though massive simulations with thousands of entities might need different settings), it will record the position of every hitbox every single physics frame. Since it's run on the server, it gets its data from your authoritative copy of the world state. Regarding deterministic results, you have complete control over how accurate Net Rewinder is. If you need more hard performance data, please let me know. Generally speaking, the raycast to determine if a bullet hits will probably be more expensive than the rewind to make sure the raycast is accurate.

In real-world use cases, you'll probably never see it above 1% of your frame time. Obviously, in a real FPS, you won't be rewinding every frame, you won't be displaying visuals for your hitbox rewinds, and you'll rarely be rewinding more than a handful of characters at a time. I've tested rewinding 125 npcs with 15 hitboxes each every frame and was able to run the simulation at 45-55FPS on my system, while rendering all 125 npcs and visualizations of their hitboxes (screenshot 2 above). You should have no trouble at all using it in a fast-paced FPS game. Works with UFPS and other FPS frameworks.Works with any network library ( UNet, Photon, Bolt, Forge, etc).Supports 2D and 3D - anything with a transform can be rewound.You can run whatever tests you want against them, and then restore them to their original positions with another call. You can use any representation of hitboxes you want, any networking library, and any hit detection routines you want. One function call rewinds hitboxes (your selected transforms) to exactly where they were at the time you specify. It focuses on doing only one thing, and doing that one thing simply and well. Net Rewinder is a server side hitbox rewinder / lag compensator for use with any networking library. Net Rewinder is now available on the Asset Store here.

The rewinder game review