# Running a test or practice session: run plans, change discipline and tyre allocation

> Track time is the scarcest resource. A run plan with clear objectives, disciplined one-at-a-time changes, controlled tyre and fuel states, and a baseline check turns a session into knowledge instead of noise.

Area: Race Engineering & Strategy · Starting level: L2 Practitioner · Updated 2026-11-16
Source page: https://race-engineer-academy.vercel.app/lesson/strategy-running-a-session

## L1 · Foundation

**Every session needs an objective.** Typical race-weekend structure:
- **FP1:** installation and systems check, track learning, baseline balance, first reading of tyre pressures and temperatures.
- **FP2/FP3:** setup direction, race simulation (long run on race fuel), qualifying preparation.
- **Qualifying:** execute the plan; no experiments.

**The run plan** lists for each run: purpose, setup change, tyre set (new/used, laps), fuel load, number of laps, what the driver should evaluate, and which data to check afterwards. Share it with mechanics (changes to prepare), the driver (what to feel), and the tyre engineer (pressures, sets).

**Change discipline:**
- **One change at a time** when you want to learn – otherwise you can't attribute the result.
- Make changes **big enough to be felt and measured** (one ARB step, not a quarter). Small changes vanish in lap-time noise (σ 0.2–0.4 s).
- **Return to baseline** at the end to check track evolution (an A–B–A sequence).
- Write down *everything*: setup, tyre set, pressures cold/hot, fuel, track temperature, driver comment, lap times.

## L2 · Practitioner

**Time budgeting.** A 60-minute session at a 1:45 track is ~25–30 laps of running if efficient. Each run costs an out-lap and in-lap; each change costs garage time. Typical turnaround times (car-dependent):

| Change | Typical time |
|---|---|
| Front/rear wing angle, brake balance (driver) | 1–5 min |
| Anti-roll bar step | 2–5 min |
| Damper clicks | 2–5 min |
| Tyre pressures | 2 min |
| Ride height / camber / toe | 10–20 min (with alignment check) |
| Springs / heave elements | 15–30 min |
| Gear ratios / diff | 30–60 min (often not possible in a session) |

Put long changes in natural breaks (red flags, between sessions) and do quick evaluations within a session.

**Tyre and fuel control.** Tyre condition is the biggest confounder. Compare setups on equal tyre age and fuel, or alternate (A on set 1, B on set 1, A again). Plan tyre allocation backwards from the race: how many new sets for qualifying, which scrubbed sets for the race start, which used sets for practice evaluation. In single-make championships with few sets per weekend (junior formulas, cup series), most evaluation runs happen on used tyres – accept it and keep the sets' history clear.

**Evaluation laps.** Two to four representative laps after tyre warm-up, in clear air, are the minimum for a judgement; long runs need 10+ laps. Use corner metrics (min speed, understeer angle, throttle acceptance) alongside lap times – they are less noisy.

**Driver briefing.** Before the run, tell the driver what changed only if it doesn't bias them – many engineers prefer *blind* evaluation of balance changes ("tell me what you feel in T3 and T7 entry") to avoid expectation bias.

## L3 · Advanced

**Designing a test day.** Treat it as an experiment:
- **Prioritise questions** by expected lap-time value × probability of learning something × cost in track time. Simulation (if correlated) can pre-screen options and leave only the uncertain ones for the track.
- **Block for drift:** track temperature and rubber change through the day; run baseline repeats (A–B–A–B) so drift doesn't masquerade as a setup effect.
- **Effect size vs noise:** if lap-time σ is 0.25 s (known from long-run history) and you can afford 4 laps per configuration, you can only resolve differences of ~0.35 s with confidence ($2\sigma\sqrt{2/n}$); if σ has to be estimated from those same few laps, use Student-t instead of 2 (t = 2.45 for 6 degrees of freedom) and it grows to ~0.43 s. For smaller effects rely on corner-level metrics with lower scatter, or on longer runs.
- **Interactions:** when two parameters are known to interact (e.g., spring change requires ride-height compensation; aero balance and mechanical balance), test them as a package, but be explicit that you test the package, not the components.
- **Sweeps:** for a single parameter with a non-linear response (e.g., rear wing, tyre pressure), test three points spanning the range to find the trend, then refine.

**Contingency planning.** Red flags, rain and mechanical issues happen. Rank the run plan so the most valuable items come first, and prepare a reduced plan ("if we lose 20 minutes, skip runs 3 and 5"). In wet sessions, decide in advance whether running provides useful information (wet setup, intermediate crossover) or only risk.

**Close the loop.** After the session: verify each change's effect with data, update the setup database with confirmed sensitivities (e.g., "+1 rear ARB step (stiffer) = −0.15° understeer angle mid-corner, −0.08 s at Paul Ricard"), and adjust the plan for the next session. Knowledge that isn't recorded is lost by the next event.

## L4 · Expert

**Power analysis: how many laps does a question need?** Significance alone ($2\sigma\sqrt{2/n}$) tells you when a measured difference is unlikely to be noise. To *reliably detect* a real effect Δ you also need power (the chance of seeing it if it is there). For two configurations, two-sided α = 5 % and 80 % power:

$$n \approx \frac{2\,(1.96 + 0.84)^2\,\sigma^2}{\Delta^2}$$

- Lap time, σ = 0.25 s, Δ = 0.2 s → n ≈ 25 laps *per configuration*. No practice session allows that.
- Minimum speed in a key corner, σ = 0.5 km/h, Δ = 1.5 km/h → n ≈ 2 laps.

This is the quantitative reason to judge small setup changes on low-noise corner metrics and to keep lap time for large changes or long runs.

**Laps are not independent.** Tyre wear, fuel burn and track evolution make consecutive laps correlated, so the effective sample size is smaller than the lap count. Two defences:
- **Blocked designs.** In an A–B–A sequence, compare B with the mean of the two A runs: a linear drift cancels exactly. A–B–B–A balances it again for the next pair.
- **Covariates.** Regress lap time on configuration plus fuel mass and tyre age, e.g. $t = t_0 + \beta_{cfg} + k_{fuel}m + d\,n_{tyre}$, instead of comparing raw means. The configuration coefficient is the effect, with an honest uncertainty.

**Small samples.** With σ estimated from the same 3 laps per side, the pooled test has 4 degrees of freedom, t = 2.78 instead of 1.96, and the resolvable difference grows by ~40 %.

**Multiple comparisons.** Test ten sensitivities at 95 % and there is a ~40 % chance ($1 - 0.95^{10}$) that at least one looks "significant" by chance. Decide before the run what each change is expected to do, and confirm surprising results with a repeat before they enter the setup database.

**Limits.** Statistics cannot fix a confounded design: if B also ran on a different tyre set or in traffic, no amount of laps rescues the comparison.

## Key formulas

**Minimum resolvable lap-time difference (95 %)**

$$\Delta t_{min} \approx 2\,\sigma\sqrt{\frac{2}{n}}$$

- \sigma: lap-time scatter (s), known from long-run history; if estimated from the same few laps replace 2 by t_{2n-2}
- n: clean laps per configuration

**Laps per configuration to detect an effect Δ (two-sided α = 5 %, power 80 %)**

$$n \approx \frac{2\,(z_{\alpha/2} + z_{\beta})^2\,\sigma^2}{\Delta^2} \approx \frac{15.7\,\sigma^2}{\Delta^2}$$

- \Delta: true effect to detect
- z_{\alpha/2}: 1.96
- z_{\beta}: 0.84

## Trackside takeaways

- Write a run plan with objective, change, tyres, fuel, laps and driver focus for every run.
- One change at a time to learn; packages only when the interaction is known and accepted.
- Changes must be large enough to exceed lap-time noise; use corner metrics for small effects.
- Control tyre age and fuel; repeat the baseline (A–B–A) to catch track evolution.
- Plan tyre allocation backwards from qualifying and race needs.

## By car category

### GT3

BoP fixes many parameters (ride height minimums, wing ranges); sessions focus on balance (ARBs, dampers, pressures) and race-pace long runs with heavy fuel.

### FIA F3

Short sessions and a limited tyre allocation: most running on used sets; quick changes (wings, bars, pressures) dominate.

### FIA F2

One practice session before qualifying: the run plan must cover quali preparation and race long-run data in very limited time.

### One-make Cup

Few adjustables and strict tyre limits: sessions are about driver development and pressures; changes should be few and clearly evaluated.

### LMP2

Endurance test days allow long runs and back-to-back aero/mechanical comparisons; driver rotation must be part of the plan.

### Super Formula

Short weekends and powerful cars with sensitive aero; quick, well-chosen changes and preparation via simulation are key.

## References

- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §2.3 Keeping Notes with Data Files; §2.8 Getting Organized
- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §15.5 How to Integrate Lap Time Simulation in Daily Data (virtual test day, test-plan screening)
- Milliken & Milliken, Race Car Vehicle Dynamics (SAE 1995): Ch. 11 Testing and Development
