# Data analysis with metrics, KPIs and run charts

> Compress thousands of samples into one number per lap. Well-designed metrics in run charts reveal trends across stints, days and seasons – and let you relate any parameter to lap time quickly.

Area: Data Acquisition & Analysis · Starting level: L2 Practitioner · Updated 2026-11-16
Source page: https://race-engineer-academy.vercel.app/lesson/daq-metrics-kpis

## L1 · Foundation

**The problem: too much data, too little time.** A modern car produces hundreds of channels per lap; between sessions you have minutes. Metrics are the answer: statistics extracted from each lap (or each corner) and plotted in **run charts** with lap number on the x-axis.

A **KPI** (key performance indicator) is a metric defined in advance that reflects success against a benchmark – like grid position averaged over a season, or "time per lap with front tyre surface temperature in its working window".

Examples of lap metrics:

| Metric | Expressed as |
|---|---|
| Average throttle position | % |
| Time at full throttle | % of lap time |
| Braking distance | % of lap distance |
| Time coasting | % of lap time |
| Max / average total brake pressure | bar |
| Average understeer angle (gated) | deg |
| Average steering smoothness | deg/s |
| Fuel used | litres or kg per lap |
| Tyre temperatures in window | % of lap time |
| Time in each quadrant of the g-g diagram | % |

Run charts show trends a single overlay can't: the car drifting into oversteer as fuel burns, fuel per lap creeping up, a driver's full-throttle % falling with tyre age. Plotting a metric against lap time (e.g., lap steering-angle integral vs lap time across a weekend) shows directly whether it matters for performance.

## L2 · Practitioner

**Building metrics correctly.**

**Gating, not zeroing.** To average lateral g in left-handers only, the naive math channel sets everything else to zero – and the zeros drag the average down. Either use software that supports signal gating (discontinuities) or compute

$$\bar{x}_{gated} = \frac{\sum_i x_i\, c_i}{\sum_i c_i}$$

with $c_i$ = 1 when the condition is true, 0 otherwise. Reset both sums at each lap beacon.

**Time and distance in condition.**

$$t_c = \int_0^{t_{lap}} c(t)\,dt, \qquad d_c = \int_0^{d_{lap}} c(s)\,ds$$

and report them as a **percentage** of lap time or distance so laps of different length/time remain comparable. Examples: time at > 98 % throttle, distance with brake pressure > 5 bar.

**Location-specific metrics.** Minimum speed in turn 4, downforce estimate between two distances on the main straight, peak brake pressure for turn 1. These are less noisy than lap-wide metrics because they isolate one phenomenon.

**Excluding junk.** Out-laps, in-laps, safety-car laps, traffic laps and laps with track-limit excursions distort metrics. Flag them (lap-time threshold, pit-limiter channel, flag status) and exclude automatically.

**Make it repeatable.** Define metrics in the analysis software as math channels and lap statistics, export the table every session, and maintain a database (car, driver, setup, tyres, fuel, conditions). The value grows with history.

## L3 · Advanced

**Thinking statistically about run charts.**

- **Control limits.** For a stable process, plot the mean ± 2–3σ of a metric (from a reference run). Points outside, or runs of 6–7 points trending in one direction, signal real change – a useful way to separate tyre degradation or a developing mechanical issue from noise.
- **Confounding.** Fuel burn, tyre age, track evolution and ambient temperature all change together through a stint. A metric that correlates with lap time over a stint may be driven by fuel load rather than causing the lap time. Use multiple regression (lap time on fuel mass, tyre age, metric) or compare at equal tyre age/fuel.
- **Normalisation.** Brake distance %, full-throttle % and coasting % are track-dependent; compare across tracks only as deltas to a reference driver or car.
- **Sensitivity metrics.** Plotting $d(\text{lap time})/d(\text{metric})$ across a weekend (e.g., 0.1 s per 1 km/h in turn 3 minimum speed) tells you where to focus engineering effort.

**Designing good KPIs.** A good KPI is (1) linked causally to lap time or reliability, (2) robust to noise (gated, location-specific where possible), (3) understood by drivers and engineers, (4) actionable – you know what to change if it moves. "Average throttle" is a performance indicator; "throttle acceptance in turn 7" is actionable.

**Examples of mature KPI sets:** tyre working-window time %, front/rear grip factor per phase (braking, mid-corner, traction), aero platform metrics (ride-height RMS at speed), driver metrics (brake aggressiveness, release smoothness, coasting %), reliability metrics (time above oil-temperature threshold). Present them on one page per run so the debrief starts with facts.

## Key formulas

**Gated average**

$$\bar{x}_{gated} = \frac{\sum_i x_i\,c_i}{\sum_i c_i}$$

- x_i: channel sample
- c_i: condition (1 true / 0 false)

**Time in condition (% of lap)**

$$T_c = \frac{100}{t_{lap}}\int_0^{t_{lap}} c(t)\,dt$$

- c(t): condition channel (0/1)
- t_{lap}: lap time (s)

## Trackside takeaways

- One number per lap in a run chart reveals trends that overlays hide.
- Gate, don’t zero – zeros bias averages.
- Express time/distance-in-condition as % of lap to compare laps.
- Exclude out/in, SC and traffic laps automatically.
- Beware confounding: fuel, tyre age and track evolution move together through a stint.

## By car category

### GT3

Multi-driver line-ups benefit hugely from per-driver metric run charts across a season (brake aggressiveness, full-throttle %, average understeer angle).

### LMP2

Endurance stints give long series of laps – ideal for run charts of fuel per lap, tyre degradation and balance drift.

### One-make Cup

Benchmark KPIs against the fastest car in the team; identical machinery makes driver metrics directly comparable.

### Formula 4

Simple metrics (full-throttle %, braking distance %, coasting %) are excellent teaching tools for young drivers.

### IndyCar

On ovals, metrics like average lift % per corner and fuel per lap at different tow conditions drive strategy decisions.

## References

- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): Ch. 17 Data Analysis Using Metrics (§17.1–17.3)
- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §2.2.5 Run Charts
