F1 PACE ANALYSIS · CASE 04 · BAHRAIN 2024 PILOT

From Practice to Race Day: Does FP2 Long-Run Pace Predict Formula 1 Race Pace?

A Bahrain 2024 methodology pilot that locks an FP2 estimate before comparing it with a separately constructed race-pace benchmark from public OpenF1 timing data.

Bahrain pilot · 20 drivers · 10 constructors · 539 retained race laps

Independent analysis using publicly available data. Not affiliated with Formula 1 or any Formula 1 team.

Predictive modelling
Race pace
Robust regression
Traffic filtering

Key results

The Bahrain validation in six measures

The locked FP2 model closely tracked the separately constructed, traffic-filtered race estimate in this pilot, while retaining meaningful error and one clear constructor-level miss.

Pearson correlation: 0.884
Spearman correlation: 0.939
Mean absolute error: 0.250 s/lap
RMSE: 0.328 s/lap
Pairwise ranking accuracy: 91.1%
Top-three overlap: 2/3

Research question

Can FP2 long-run pace predict race-day competitive order?

This project tests whether adjusted FP2 long-run pace predicts adjusted, traffic-filtered race pace and competitive order. The FP2 prediction was locked before the race sample was constructed, keeping the race analysis as a separate validation rather than allowing the race result to reshape the FP2 model. This mirrors an early-weekend strategy task: form a pre-race view from incomplete practice data, then test it against later evidence without back-fitting.

Data and cleaning pipeline

Two independently audited pace samples

Public timing feeds were filtered into one representative FP2 long run per driver and a separate, traffic-filtered race sample, with each exclusion preserved in the audit trail.

Source data OpenF1 supplied lap, stint, tyre, weather, race-control, position and interval data
FP2 sample One genuine FP2 long run selected for each of the 20 drivers
Lap audit Pit laps, incomplete laps, interruptions, anomalously slow laps and observable traffic audited, with each exclusion reason recorded
Race traffic A 2.0-second exclusion threshold applied using the official interval-to-car-ahead feed
Pace adjustment Robust regression accounted for tyre compound, tyre age, track temperature and session progression
Final race model The final race model retained 539 laps across 39 stints, 20 drivers and 10 constructors

Twenty small-multiple charts showing each Bahrain 2024 FP2 driver’s selected long run, retained pace laps, removed slow anomalies and final Huber tyre-age trend. — The 20-panel audit keeps identical axes for every driver and distinguishes retained pace laps from slow anomalies removed before the final robust tyre-age trend was fitted. Select the chart to inspect the full-resolution figure.

Adjusted FP2 prediction

Red Bull led the locked FP2 estimate

Negative values indicate faster-than-field-median adjusted pace. The FP2 model predicted Red Bull first, followed by McLaren and Mercedes.

Horizontal bar chart of adjusted Bahrain 2024 FP2 constructor gaps to the field median, with Red Bull fastest ahead of McLaren and Mercedes. — Locked adjusted FP2 constructor pace gaps relative to the field median; negative values represent faster estimated pace. Select the chart to inspect the full-resolution figure.

Adjusted traffic-filtered race pace

The independent race model moved Ferrari to second

The independently constructed race model ranked Red Bull first, Ferrari second, Mercedes third and McLaren fourth.

Horizontal bar chart of adjusted, traffic-filtered Bahrain 2024 race-pace gaps to the field median, ranking Red Bull first, Ferrari second, Mercedes third and McLaren fourth. — Adjusted constructor pace gaps from the independently filtered race sample; negative values represent faster estimated pace. The chart uses “clean pace” as shorthand for this adjusted, traffic-filtered estimate. Select the chart to inspect the full-resolution figure.

Prediction validation

Strong ordering signal, with Ferrari as the largest miss

The validation supports a strong Bahrain relationship between the locked FP2 prediction and traffic-filtered race pace, but it does not show that FP2 predicted the race perfectly.

Fastest constructor identified

FP2 identified Red Bull as the fastest constructor.
Largest miss

Ferrari was predicted fourth in FP2 but ranked second in traffic-filtered race pace.
Rank ordering largely retained

Eight of ten constructors finished within one rank of the FP2 order; Ferrari and McLaren each moved two positions.
Pairwise accuracy

FP2 correctly ordered 41 of 45 constructor pairings, or 91.1%.

Scatterplot comparing locked adjusted Bahrain 2024 FP2 constructor gaps with adjusted, traffic-filtered race gaps, showing broad agreement around the diagonal and Ferrari as the largest miss. — Each point compares one constructor’s adjusted FP2 gap with its adjusted, traffic-filtered race gap. The chart uses “clean race pace” as shorthand for the traffic-filtered race estimate. The diagonal marks equal FP2 and race estimates; distance from it shows prediction error. Select the chart to inspect the full-resolution figure.

Robustness

The ordering remained stable under a less conservative traffic filter

Lowering the race-traffic exclusion threshold from 2.0 to 1.5 seconds applied a less conservative filter and retained more laps. The main competitive ordering and validation metrics remained close to the primary estimate.

Retained laps: 539 → 603Increase under the less conservative 1.5-second filter.
Smallest driver sample: 5 → 13Minimum retained laps across drivers.
Race-order stability: 0.988Spearman correlation versus the primary race estimate.
Validation Spearman: 0.939 → 0.927Primary estimate versus the result under the less conservative filter.
Mean absolute error: 0.250 → 0.270 s/lapPrimary estimate versus the result under the less conservative filter.

Limitations

What this pilot cannot establish

The result is a relative constructor-level validation for one event, not a general claim that practice pace will predict every race.

FP2 fuel loads are unobservable.
Fuel burn and tyre degradation cannot be perfectly separated.
FP2 traffic detection relies on approximate location data.
The validation contains only 10 constructor-level observations from Bahrain 2024, so its correlations describe this event rather than general predictive reliability.
Driver pace management and strategy effects remain only partly observable in public timing data.
Bahrain’s closely matched evening FP2 and race conditions may make it unusually favourable.
The outputs are relative constructor gaps, not predictions of absolute lap time.

Next phase

Test the fixed method at two more races

The fixed methodology will next be tested unchanged at Spain and Abu Dhabi before expansion to a larger eligible 2024 sample. FP1 and FP3 remain later sensitivity comparisons rather than part of the primary predictor.