PP Is the Most Misleading Number in CSR2

PP is the number CSR2 puts in front of you most often. It is on the car select screen, it goes up when you fit a part, and it is the figure players quote at each other when they argue about builds. It is also, in the tier where the game is actually played, close to useless as a predictor of who wins.

We tested it. For every tier we took all the cars with both a PP figure and a recorded time and measured how strongly PP tracks real elapsed time. 1,251 cars have a recorded time; PP is present on all 1,309.

PP gets worse as the game gets serious

A correlation of -1.0 would mean higher PP always means a quicker time. 0.0 would mean PP tells you nothing at all.

Read the last column carefully. In Tier 5, if you pick any two cars at random, there is a 31.8% chance the quicker one has the lower PP. Nearly a third of the time the number on the screen points the wrong way.

Tier 1 is the extreme case. With 16 cars the correlation is +0.06, meaning PP has no relationship with elapsed time whatsoever, and the sign is technically the wrong way round. We would not read much into 16 cars, but it is certainly not evidence that PP works.

At the top, PP stops varying at all

Here is the finding that explains why experienced players ignore it. Take the 25 fastest cars in Tier 5 and look at their PP figures. Every single one is 713. Not clustered around 713 — exactly 713, all twenty-five.

The entire fight at the top of CSR2 happens between cars with identical PP. The number has no resolving power there at all, because it has stopped varying while the times have not.

Meanwhile the highest-PP car in the game is Letty's Chevrolet Nova SS from the Fast and Furious set, at 726 PP. It records 7.004 and sits 268th in its tier. It carries 13 more PP than every car in the top 25 and finishes 0.499 seconds behind the quickest of them, and 0.352 behind the slowest.

EVO is the number you should be reading

We ran the same test against EVO instead of PP, on the same Tier 5 cars. PP scores -0.20. EVO scores -0.96.

That is not a small improvement, it is a different class of number. EVO tracks real recorded time almost exactly across the Tier 5 field, where PP barely tracks it at all. If you want one figure on the car select screen to judge a build by, it is EVO, and it is not close.

Put in plainer terms: across those 890 Tier 5 cars, PP accounts for roughly four per cent of the variation in recorded times. EVO accounts for about ninety-two. That is the whole case against reading the PP figure, and it is not that PP is random. It is that PP stops discriminating between cars at exactly the point where you need it to, while EVO keeps going.

This also explains a common and frustrating experience. You fit a part, PP jumps, and the car feels no quicker. That is not your imagination and it is not a bug. PP counts what you have fitted; it does not weigh those parts by how much time they actually save on a run.

But EVO breaks down at the sharp end too

Here is where it would be easy to oversell the conclusion, so we will not. A correlation of -0.96 across 890 cars is excellent, and it still does not make EVO a reliable tiebreaker between two fast cars.

The clearest counterexample is the top of the game. The BTTF Part III Time Machine records 2,699 EVO, the highest figure in CSR2. The Honda Hoonigan Indy Truck records 2,525, a full 174 points behind. The truck loses by six thousandths of a second.

That is the correct way to hold both facts at once. EVO is a very good predictor of roughly where a car will land in the field. It is a poor predictor of which of two already-fast cars will win, because at that end the times are compressed into hundredths while EVO is still spread across hundreds of points.

What to actually do with this

• Stop comparing cars by PP. In Tier 4 and Tier 5 it points the wrong way roughly a third of the time.
• Use EVO for rough sorting. It is a strong guide to where a car sits in the broad field.
• Use recorded times for any decision that matters. If two cars are close, only the times separate them.
• Never compare PP across tiers. Tier 5 runs 642 to 726 and Tier 4 runs 546 to 676, and the race distances differ, so the comparison is meaningless in both directions.
• Treat a high-PP, low-rank car as a warning. The Nova SS is the clearest example in the game of PP flattering a build.

None of this means the game is lying to you. PP is doing what it was designed to do, which is describe how much performance has been fitted to a car. The mistake is reading a description of the build as a prediction of the result. Those are different questions, and on this dataset they give different answers about a third of the time.

Every figure here is reproducible from the record database. Filter by tier, sort by time, and read the PP column down the page — the pattern is visible without any statistics at all.