int_lap_air_state

Air state classification and thermal load from dirty air.

Part of the Physics family.

Air state classification and thermal load from dirty air. One row per stint-lap. Classifies air state per sector (free_air, tow_zone, drs_train, dirty_air) and accumulates cumulative thermal load on tyre surface and bulk using exponential moving average. S2 (the dirty-air third) is classified on the median gap alone: < 1.5 s is dirty_air whatever DRS did (F48, WI-15a). drs_train and tow_zone arise only on the straight thirds (S1/S3), where DRS still splits a sub-1 s follow. FD6 (CO-04): a lap with no measured gap to the car ahead is ‘free_air’ / 0.0 only for the race leader (end-of-lap position 1), who has no car ahead. For any other car, or an unknown position, every air-state term (air_state_dominant, dirty_air_share_lap, tow_benefit_lap_s, dirty_air_intensity, time_in_dirty_air_s and the thermal loads) is NULL: unmeasured, not clean air. The NULL wins over the SC/VSC zeroing. A thermal load is also NULL while an unmeasured lap of the same stint is inside its lookback. 2018_1 Australia and 2018_2 Bahrain have no telemetry in bronze, so nearly all of their non-leader laps are NULL here; downstream that excludes both races from the field base (int_field_pace_curve) and loses their residuals and training labels. Elsewhere this is 2-32 laps a season. The dominant state’s tie-break is declared (T7): most frequent measured sector state, ties to dirty_air > drs_train > tow_zone > free_air.

​

Lineage

​

Overview

Property	Value
Layer	intermediate
Family	Physics
Contract enforced	No
Upstream	int_stint_geometry, stg_laps

12 tests guard this model - 5 generic, 7 singular.

​

Columns

Column	Type	Nullable	Tests	Description
lap_id		Yes	not_null, unique	Foreign key to stg_laps
dirty_air_share_lap		Yes		1.0 when the lap’s S2 median gap to the car ahead is under 1.5 s, else 0.0 (one S2 per lap, so it is a bit, not a fraction). Zeroed on SC/VSC/red-flag laps. Non-increasing in s2_gap_median_s by construction since F48 (WI-15a): before it, a sub-1 s S2 gap with DRS open was coded drs_train and billed 0 while 1.0-1.5 s was billed 1. Asserted by assert_dirty_air_share_monotone (T38). A contract feature, and lagged one lap it is the treatment in int_dirty_air_tax_component’s theta_air, so it moves the label.
s2_gap_median_s		Yes	expect_column_values_to_be_between	F48 (WI-15a). The measured S2 median gap to the car ahead (seconds; DistanceToDriverAhead / point speed, median over the third’s samples), i.e. the quantity dirty_air_share_lap is a function of. Raw: not zeroed on SC/VSC laps. NULL when the lap has no S2 gap sample. Not in the ML contract.
s2_drs_active		Yes		F48 (WI-15a). TRUE when any S2 sample had DRS open (FastF1 codes 10/12/14). This used to be folded into the S2 classification (it turned a sub-1 s dirty-air lap into a clean one); it is carried here instead, so the unsettled physics question (do DRS-open S2 laps lose less aero load?) stays answerable without the label assuming an answer. NULL when the lap has no S2 telemetry. Not in the ML contract.
time_in_dirty_air_s		Yes	not_null, expect_column_values_to_be_between	Real per-lap dirty-air exposure time (seconds), summed from actual ~10Hz telemetry sample durations where gap-to-car-ahead < 1.5s. Zeroed on SC/VSC laps (no real aero load at safety-car speeds). Null-gap samples within a measured lap are not counted. A lap with no gap sample at all is 0.0 for the race leader and NULL for anyone else (FD6, CO-04), so not_null holds only where min_gap_s is measured.