What makes a test discriminating
A test is useful when its result tells two explanations apart. If a driving and a setup explanation predict the same outcome, the run is spent and nothing is learned.
| Explanation | Hold the setup, repeat run A's driving cue | Revert the setup, keep run B's driving |
|---|---|---|
| Driving: earlier lift, later throttle | The corner-window gap closes. | The gap stays until the driving changes. |
| Setup: front roll bar 3 to 4 | The gap stays. | The gap closes. |
The five parts of a test plan
Every plan needs these five parts. The test plan Racing saves uses the same fields.
| Part | Question it answers | Example |
|---|---|---|
| One change | What differs from the last run? | Repeat run A's lift and throttle cue through the corner. Keep the setup. |
| Hold constant | What must not change? | Front roll bar at 4, fuel within 2 L, the same tyre set, the same driver. |
| Measurement | What will you read? | Corner-window minimum speed, the throttle return point and the median over clean laps. |
| Success criterion | What result separates the explanations? | The corner-window gap to run A closes on clean laps. |
| Stop condition | When do you stop? | After three clean laps, or at the first sign the tyres have gone off. |
Choose the test that costs the least
Track time is scarce. When two tests discriminate equally well, run the one that needs no setup change: it is quicker, cheaper and easier to repeat. That is why a driving test often comes first.
Hold the conditions honestly
- Keep fuel within a narrow band, or state the difference.
- Use the same tyre set, or tyres of similar age.
- Run close in time to the baseline, so the track is in a similar state.
- Keep the same driver, unless the driver is the variable.
Record the result, even when it is dull
A test that shows no difference still rules something out. Record the result against the success criterion, keep the plan with the run and choose the next test from what is left.
Common mistakes
- Changing two things at once, so neither can be credited.
- Deciding the success criterion after seeing the data.
- Too few laps to separate the change from lap-to-lap spread.
- No stop condition, so a failing test uses the whole session.