Weigh driving against setup, with the evidence.
Racing's AI reads the synthetic Jerez comparison and writes competing driving and setup hypotheses, each tied to the evidence it cites and to what is missing. You correct the context, request a revised analysis and leave with one editable test plan. It advises; engineering judgment stays with you.
Engineer’s debrief
AI-generated, advisoryEvidence, competing explanations, and a test to separate them.
Driving explanation
Run B lifts earlier and returns to throttle later through the corner window, which may explain the slower exit.
- B02 corner window
- A02 corner window
Setup explanation
The front roll bar moved from position 3 to 4 in run B, which may add entry understeer. These traces alone cannot establish the cause.
- Run B context
- Run A context
What remains uncertain
Hot tyre pressures were not measured in either run. Adding them as context would narrow the setup explanation.
Missing evidence
- Hot tyre pressures, run A
- Hot tyre pressures, run B
Proposed next test
Repeat before changing the car
Hold the setup fixed. Repeat run A's lift and throttle cue on clean laps. Compare the corner window and run medians.
Run B against run A
Jerez · Fictional GT4 #27 · Speed (km/h) · Lap distance (m), sector 2
- Median lap delta
- +0.780s
- A02 / B02 · sector 2
- +0.580s
Reading the chart
- Run A, A02
- Reference lap from run A. Front roll bar at position 3.
- Run B, B02
- Compared lap from run B. Front roll bar moved to position 4.
- Corner window
- 2100 to 2800 m. The window the evidence cites.
View speed data as a table
| Distance (m) | Run A, A02 (km/h) | Run B, B02 (km/h) |
|---|---|---|
| 2000 | 166 | 166 |
| 2100 | 166 | 166 |
| 2200 | 166 | 163 |
| 2300 | 131 | 127 |
| 2400 | 75 | 73 |
| 2500 | 75 | 73 |
| 2600 | 131 | 127 |
| 2700 | 167 | 163 |
| 2800 | 167 | 167 |
What the AI race engineer does today
Everything here ships today on the synthetic scenario. The model runs on the OpenAI API and is called only when you request an analysis.
- Reads the evidence you selected: run medians, the laps you kept, the A02 and B02 corner windows and each run's reported context.
- Writes driving and setup hypotheses, each with the evidence it cites, what remains uncertain and which measurements are missing.
- Proposes one next test that can tell the explanations apart.
- Takes your corrections. Add what the data cannot show, then request a revised analysis. Earlier results keep the revision they were built on.
- Turns the proposal into one editable test plan you save: the change, what to hold constant, the measurement, the success criterion and the stop condition.
What it does not do
It interprets measured differences; it does not prove a cause. Every result carries the label AI-generated, advisory.
- It does not fill a missing measurement with a guess or a zero.
- It does not claim a lap-time gain for any change.
- It does not choose the test for you. The plan is yours to edit before you save it.
What is sent, and when
Only when you request an analysis, the selected evidence and your saved context notes go to the OpenAI API with storage disabled. The provider may keep requests for up to 30 days for abuse monitoring. Data ownership shows the full flow.
The method behind the debrief
The debrief follows the order an engineer would. Driving or setup: separating causes explains how to read the traces. Designing one discriminating test explains how the next test is chosen.
Design-partner roadmap, 2027 season
Planned with design-partner teams for the 2027 season. Not available today.
- A per-team isolated model trained only on that team's data. No cross-team training and no shared model.
- MoTeC and AiM CSV import, so the debrief reads your own logged sessions.
- Team data stays the team's, with export and deletion when a team leaves.
Read the debrief on your own screen.
Private early access. Request an analysis of the synthetic Jerez comparison and inspect every citation.
Request access