Netcode and sim racing incidents, or why you both swear the other one hit you

Everyone who races online has had this one. You're fully alongside, the other car doesn't seem to see you, and you both end up in the wall. On your screen they turned into you. On theirs, you drifted over. Both clips look clean, and both drivers are telling the truth.

This is netcode, and it's the least discussed reason why sim racing incidents are so messy.

Everyone is racing a slightly different race

When you drive online, you don't see the other cars where they are. You see where they were a short time ago, plus a guess about where they are now. Every car's position has to travel over the internet, which takes time, and your sim fills the gap by predicting.

So on your screen the other car is a bit behind its real position, and on their screen yours is too. How much depends on each connection, and it changes second to second.

At 200 km/h, a car covers about 55 metres a second. A delay of just 50 milliseconds is almost three metres, or most of a car length. You don't have to be on a terrible connection for this to matter.

Prediction is mostly good, until it isn't

The sim makes smart guesses: if a car was going straight at a certain speed, it'll probably keep doing so. This works great on straights and badly in the exact moments that cause incidents: braking, turning in, and fast direction changes. The predicted position and the real one drift apart most when a car's input changes suddenly.

That's why close side-by-side racing is where the problems show up. The cars are within centimetres of each other, and the errors are measured in centimetres too.

Who decides if there was contact?

It depends on the sim, and on how the session is hosted. Broadly, there are a few approaches:

  • A central server works out the "truth" and tells everybody. Both players might see something slightly different, but the server's version is the official one.
  • Some designs let each driver's machine decide what happens to their own car. Then you can get contact on one screen and none on the other.
  • Some blend the two.

You don't need to know which one your sim uses. You need to know that in all of them, the replay you see may not be what the other driver saw, and in some it may not even be what the server believes.

What this does to stewarding

A few patterns come up again and again:

Phantom contact. One driver's clip shows a hit, the other's shows nothing. The rule "we penalise what we can see" is fine, but remember that "what we can see" depends on whose clip it is.

The squeeze that wasn't. A car appears to move across to close the gap, when in fact the other car's position just updated suddenly. The driver reporting it is genuinely certain.

The ghost brake. The car ahead looks like it slammed on the brakes, and the driver behind couldn't avoid it. Usually the car did brake, just a moment earlier than it looked. Rear-end incidents are especially hard to judge for this reason. See incident examples.

The one with the bad connection. Drivers with high ping tend to be the ones who suddenly "teleport". They often don't feel the problem themselves and are blamed for the results.

What leagues can do about it

None of this makes online racing unfixable. A few habits make it much easier to live with:

  1. Ask for clips from both drivers. Not just the reporter. If both agree on what happened, you're on firm ground.
  2. Get replay data from the host if you can. If your sim saves a server-side replay or telemetry, treat it as better evidence than any single driver's view.
  3. Write a connection rule. Something like "if either driver had visible lag in the incident, the stewards may call it a racing incident." Drivers accept it much more easily if it was written beforehand.
  4. Be more careful with close contact. The more precise the verdict depends on a few centimetres, the less you should trust any individual clip.
  5. Keep a record of connection issues. If someone is repeatedly involved in "weird" incidents, check whether it's their connection. Support them in fixing it before blaming their driving.
  6. Treat repeated patterns differently from one-offs. A single phantom hit is bad luck. Five in a season from the same driver is worth a conversation.

What this means for AI review

Any tool that reads clips, including ours, is bound by the same limit: it sees one driver's version. A good report should say so, flag low confidence when the angles don't agree, and let the steward decide. We wrote more about this in what AI can and can't judge.

If you take one thing from this: when two honest drivers disagree, the first question isn't "who's lying?" It's "what did each of them actually see?"