We dropped a car’s reliability from 99 to 60 and it finishes in the same place

Every team in our simulator has a number called reliability. It runs from 0 to 100 and anyone would read it as exactly what it looks like: how well that car holds together. We took the 2026 grid, dropped that number from 99 to 60 without touching anything else and ran 60 full championships at each step. The car that never breaks and the car that breaks constantly finish in the same place. Reliability, in our engine, barely exists.

The sweep: 99, 90, 80, 70, 60

Three teams from different parts of the grid, because a car already running last has no places left to lose and a null result there would prove nothing. Only one thing changes in each row: the reliability of the team being measured. Rivals, drivers, car performance and even the random seeds are identical across all five runs, so anything that moves is down to reliability.

We covered two thirds of the whole scale and the front-running team loses 0.28 championship places. Not two places: less than a third of one. The midfield team loses 0.18 and the backmarker 0.37. In points, the big team gives up 3.6% going from an indestructible car to one falling apart.

For comparison: a single point of car performance moves more than those thirty-nine points of reliability. The reason is in the engine's retirement formula, where reliability carries a weight of 0.04. At that weight, the number can travel its entire scale without the result noticing.

And the consequence shows up in the predictions

This is not a laboratory curiosity. It leaves a measurable mark on the one thing we publish in advance. Across the 4 Grands Prix we predicted before they ran, the model expected 4.0% of cars to retire. The real figure was 18.2%: 16 of 88 starts.

That is 4.6 times more broken cars than the simulator counted on. And it is not that 2026 is an unusual year at both ends: the engine simply breaks few cars, always, whatever sits in that box.

Why it is like this and why we have not touched it yet

The honest answer is that the whole engine was calibrated with this behaviour. Car performance figures, driver ratings and the balance of all eleven game modes are tuned against a world where almost nobody retires. Raising the weight of reliability would fix this one box and knock everything else out at the same time: seasons would stop reproducing, saved player records would sit on a different scale, and the calibration would have to be redone from scratch.

There is a second reason, and this one is a deeper problem we found while measuring something else. When we tried to calibrate driver ratings by comparing a driver's real championship position against the simulated one, the method kept asking us to lower the rating of drivers in fragile cars: championship position has retirements baked into it, and since the engine does not reproduce them, what the method really did was push the car's fragility into the driver's rating. A driver who retired five times looked like a worse driver. We stopped it, and that comparison is now made on average finishing position when he finishes.

So we are writing it down somewhere it can be read: when you see a car's reliability on this site, it is a historical figure we display, not a lever that moves the result. Everything else the simulator shows does move it. The rest of the model's known limits are in how the simulator works under the hood and in the seasons it cannot explain. And if you want the opposite question, how often it gets things right when it commits before a race, that is in the simulator's scoreboard.