The replay session: finding the mistake you cannot feel

The mistakes you can feel are already fixed; the ones costing you time are the ones that feel fine. A replay and four telemetry channels find them in twenty minutes, and no amount of additional driving will.

Why feel is not enough

Feel reports what changed, not what was optimal. A corner taken two km/h below the limit feels identical to one taken at it, because nothing unusual happened, and the lost tenth is invisible from inside the car. This is why drivers plateau: they have fixed everything that felt wrong.

The force feedback has the same limit. The published readouts include a force feedback meter where red indicates clipping, meaning the wheel capacity has been exceeded, and a driver running in clipping is being told less than the simulation is sending. Checking that meter is the first step of any analysis, because a clipping wheel invalidates every conclusion drawn from feel.

And the rest of the meter box is about the machine rather than the driving. The published thresholds are specific: full bar on the physics thread meter means about 16.7 ms at 60 Hz, full bar on latency means one second, full bar on packet loss means 50 per cent per direction, and full bar on page faults means 30 per second. Rule those out before diagnosing yourself.

The four channels to read first

Speed against distance is the first, and it is the only one that shows where the time went. Overlay your fast lap and your average lap and look for where the lines separate; the gap almost always opens before the corner where it becomes visible.

Throttle against distance is the second. Look for the point of full throttle at each corner exit and compare it between laps. Getting to full throttle ten metres earlier at a corner leading onto a long straight is worth more than anything you can do at the entry.

Brake against distance is the third, and here you are looking at the shape rather than the peak. A brake trace that reaches its maximum instantly and holds is a straight line braking trace; one that peaks and then falls away gradually is trail braking. Which one you are producing is not always what you think.

Steering angle against distance is the fourth and the most diagnostic. Adding angle without the car rotating is understeer, visible as a plateau in yaw against a rise in steering; corrections show as a sawtooth. A smooth single arc through a corner is what you are aiming for, and comparing your trace with a faster driver's is the fastest way to see the difference.

Turning a replay into a change

Watch each incident from three camera angles: yours, the other car's and an outside view. The rulebook is explicit that the software does not attempt to determine fault and assesses incidents to everybody involved, so the useful question is never who was wrong but what you could have done differently with the information you had.

Write down one change per session. Not five, because five changes cannot be tested, and the point of the analysis is to produce something you can verify next time out.

And re watch a race you thought went well. Errors in a good race are the expensive ones, because nothing drew your attention to them and they will repeat. A race that ended in a crash teaches you one thing; a clean race with four small mistakes teaches you four.

The twenty minute routine

Five minutes on the meters to rule out clipping, latency and frame time. Five minutes overlaying speed against distance to find the two corners costing the most. Five minutes on throttle and brake traces at those two corners. Five minutes writing down one change and the lap time you expect it to produce.

The arithmetic favours the routine heavily. 20 minutes of analysis against a 40 lap race of about 60 minutes is a 1 to 3 ratio, and finding 2 tenths in each of the 2 worst corners is 4 tenths a lap, or 16 seconds over those 40 laps. No 20 minutes of extra driving has ever returned 16 seconds.

Then test it in the next session and check whether the prediction was right. An analysis that does not produce a testable prediction is entertainment rather than work.

Sources

1. iRacing, in game meter box and telemetry readouts
2. iRacing, official sporting code on incident assessment

Frequently asked questions

Which telemetry channel matters most?

Speed against distance, because it is the only one that shows where the time was lost. The others explain why.

Why check the meters before the driving?

Because a clipping wheel or a stuttering physics thread invalidates every conclusion you draw from feel. Rule out the machine first.

Should I analyse races I lost or races I won?

Both, but the clean races are more valuable. Nothing drew your attention to the small mistakes, so they repeat.

How many changes should come out of one session?

One. Five changes cannot be tested, and an untested change is a guess wearing the clothes of analysis.