Race pace

Clean the laps, with an audit

LapCleaningPipeline drops laps unsuitable for analysis and reports how many each step removed:

import fastf1
from lapbox.pace import LapCleaningPipeline

session = fastf1.get_session(2025, "Monza", "R")
session.load(telemetry=False, weather=False, messages=False)

cleaning = LapCleaningPipeline().clean(session.laps)
print(cleaning.initial_rows, "->", cleaning.final_rows)
print(cleaning.removed)

974 -> 916
{'duplicates': 0, 'missing_laptime': 0, 'pit_laps': 41, 'inaccurate': 17, 'deleted': 0, 'implausible_time': 0, 'outliers': 0}

Each threshold is a field of LapCleaningConfig. The outlier step is off by default; turn it on with a robust z-score threshold:

from lapbox.pace import LapCleaningConfig

strict = LapCleaningPipeline(LapCleaningConfig(outlier_mad_threshold=3.5)).clean(session.laps)
print(strict.final_rows, strict.removed["outliers"])

888 28

The ranking functions below do their own filtering, so they take session.laps directly. Clean first for TyreDegradationModel and for your own analysis.

A circuit’s normal pace

from lapbox.pace import representative_base_time

print(round(representative_base_time(cleaning.frame["LapTimeSeconds"]), 3))

83.552

representative_base_time is the median of the laps within 107 % of the best, ignoring rain, safety-car and traffic laps. With fewer than 20 such laps it falls back to the plain median: a handful of laps is a worse estimate than a contaminated many.

Take the fuel out

fuel_correct normalises every timed, non-pit lap to end-of-race fuel, taking off 0.048 s for every lap still to run:

from lapbox.pace import fuel_correct

corrected = fuel_correct(session.laps)
print(corrected[corrected["Driver"] == "VER"].head(3).round(3).to_string(index=False))

Driver  LapNumber Compound    raw  corrected
   VER          1   MEDIUM 87.159     84.663
   VER          2   MEDIUM 84.859     82.411
   VER          3   MEDIUM 83.512     81.112

It returns lap 1 too; the rankings below leave it out. The race distance defaults to the highest lap number in the laps — pass total_laps= if the laps you pass don’t reach the end.

True pace

true_pace_ranking ranks each driver’s median over representative laps — from lap 2, no pit laps, within 107 % of their best — raw and fuel-corrected:

from lapbox.pace import true_pace_ranking

pace = true_pace_ranking(session.laps)
print(pace[["driver", "laps", "raw_median", "corrected_median", "gap", "rank_change"]].head(5).round(3).to_string(index=False))

driver  laps  raw_median  corrected_median   gap  rank_change
   VER    50      82.874            81.215 0.000            0
   NOR    50      82.966            81.560 0.345            1
   PIA    50      82.938            81.651 0.436           -1
   LEC    50      83.102            81.803 0.588            0
   RUS    50      83.140            81.892 0.677            0

A positive rank_change means a driver moves up once fuel is accounted for: their quick laps came on heavy fuel. Norris and Piastri swap here, and the corrected top five matches the finishing order — one race, which proves nothing about the method.

Clean air against traffic

A car within about two seconds of the one ahead loses time in its wake. clean_air_ranking splits each driver’s fuel-corrected laps by the gap to the car ahead at the line and compares the two medians:

from lapbox.pace import clean_air_ranking, clean_air_split

print(clean_air_ranking(session.laps)[["driver", "clean_laps", "traffic_laps", "delta"]].head(3).round(3).to_string(index=False))
law = clean_air_split(session.laps, "LAW")
print(f"clean {law.clean_pace:.3f} s, traffic {law.traffic_pace:.3f} s, led {law.led_laps} laps")

driver  clean_laps  traffic_laps  delta
   LAW          10            40  0.647
   ALB          24            26  0.542
   PIA          42             8  0.492
clean 82.109 s, traffic 82.756 s, led 0 laps

Lawson spent 40 of his 50 counted laps in traffic and was 0.647 s a lap slower there. The 2 s threshold (DIRTY_AIR_S) is a rule of thumb, not a measurement; change it with threshold=:

print(clean_air_split(session.laps, "LAW", threshold=1.0).traffic_laps)

35

A driver needs three laps on each side to be split; otherwise clean_air_split returns None. Gaps only count cars on the same lap, so a lapped car in front is not traffic.

Consistency

from lapbox.pace import consistency_ranking, driver_consistency

print(consistency_ranking(session.laps).head(3).round(3).to_string(index=False))
ver = driver_consistency(session.laps, "VER")
print(ver.laps, round(ver.std, 3), [(lap, round(off, 3)) for lap, off in ver.series[:3]])

driver  laps   best   mean   std    cv
   ALO    20 83.757 84.539 0.485 0.574
   OCO    49 82.892 83.977 0.664 0.791
   STR    49 82.164 83.972 0.748 0.891
50 0.853 [(2, 3.856), (3, 2.509), (4, 2.259)]

The spread of representative laps (from lap 2, no pit laps, within 107 % of the best), most consistent first by coefficient of variation. series holds every counted lap as seconds off the driver’s best — ready to plot.

The ideal lap

ideal_lap adds up a driver’s best three sectors and compares them with their fastest lap; biggest_loss says where that lap left the most time:

from lapbox.pace import biggest_loss, ideal_lap

ideal = ideal_lap(session.laps, "VER")
print(f"ideal {ideal.ideal:.3f} s, fastest {ideal.fastest:.3f} s (lap {ideal.fastest_lap}), gain {ideal.gain:.3f} s")
worst = biggest_loss(ideal)
print(f"most to find in S{worst.sector}: {worst.loss_on_fastest:.3f} s (best on lap {worst.lap})")

ideal 80.974 s, fastest 81.003 s (lap 52), gain 0.029 s
most to find in S1: 0.029 s (best on lap 42)

The ideal is rarely achievable in one lap: the best sectors usually come from laps on different fuel and tyres. It returns None rather than a misleading number when a sector was never timed on a clean lap, or when the sum comes out slower than the fastest lap. It works on any session; in qualifying it is the classic “ultimate lap”.

Next: Tyre degradation.