int_dirty_air_tax_component

One row per lap.

Part of the Physics family.

For the conceptual foundation, read Int Dirty Air Tax Component.

One row per lap. Cost of driving in dirty air (turbulent wake) following another car. Uses prior-lap dirty-air intensity to establish causality: current lap slowdown → prior lap’s position. Avoids reverse causality. Seconds lost = θ_air × dirty_air_intensity_lag1, capped [0, 5.0] seconds. θ_air is the DECLARED var theta_air_label_value (0.331 s/lap since W40, 2026-09-27), applied as-is. It is fitted from the data, but outside dbt: 06b’s F2 estimator (partial residual on the lagged dirty-air bit, stint fixed effects plus six tyre-age bins) on this model’s two-arm calibration panel, 2018-2025, gives 0.3313 [0.293, 0.370] (_roadmap/_fixes/wi/WI-12-06b-remeasure.md). The pooled OLS slope this model used to fit in SQL (0.5033 on the same panel, WI-01’s interim value) billed some car pace as dirty air, because slower cars follow more often. History: an unfitted 0.5 literal through v12 (the panel was filtered to the treated arm, so the slope was unidentified, 08q), 0.1310 (08q, v12 substrate), 0.5033 (WI-01, interim), 0.331 (W40). One θ for every season; per-season fits are published in the 06b post, not carried into the label.

​

Lineage

​

Overview

Property	Value
Layer	intermediate
Family	Physics
Contract enforced	No
Tags	causal_decomposition, dirty_air
Upstream	int_lap_fuel_state, int_field_pace_curve, int_compound_cliff_predicted, int_stint_geometry, int_lap_air_state, int_event_corrections, int_track_evolution

13 tests guard this model - 8 generic, 5 singular.

​

Columns

Column	Type	Nullable	Tests	Description
lap_id		Yes	not_null, unique	PK, FK to stg_laps
dirty_air_intensity_lag1		Yes		One-lap-lagged dirty-air intensity (dirty_air_share_lap) from int_lap_air_state, over the stint’s full lap sequence; 0.0 on the stint’s first lap. NULL when the previous lap’s air state is unmeasured (FD6 / CO-04: no gap and not the leader), not 0.0. Asserted by assert_dirty_air_tax_null_only_when_unmeasured.
dirty_air_tax_s		Yes	not_null, expect_column_values_to_be_between	Per-lap dirty-air slowdown in seconds. θ_air × dirty_air_intensity_lag1. Positive (slower cost). Bounded [0, 5.0]. θ_air is the declared theta_air_label_value (0.331 s/lap, W40), fitted outside dbt on the two-arm calibration panel (see the model description). Because dirty_air_intensity_lag1 is one bit per lap, this column takes two values per build — 0.0 on a clean-air lap and θ_air on a dirty-air lap. A dose-response on the measured gap (int_lap_air_state.min_gap_s / stg_telemetry_position.distance_to_ahead_m) is the standing follow-up; removing the treated-arm filter is the identification fix, not the measurement fix. NULL exactly where dirty_air_intensity_lag1 is (FD6 / CO-04).
dirty_air_tax_se_s		Yes	not_null	Standard error of the tax estimate via propagation. NULL where the tax is.
tax_calibration_confidence		Yes	not_null, expect_column_values_to_be_between	Confidence in the tax [0, 1], as n / (n + 500) over the calibration sample n. Since 08q restored both arms to that panel, n is the full two-arm panel (138,970 laps on the WI-01 label) rather than the 28,523 treated laps it was through v12, so this reads ~0.996 rather than ~0.983. It measures calibration SAMPLE SIZE ONLY and is not a statement about whether θ_air is distinguishable from zero in a given season — it is near 1.0 on every season including 2021 and 2024, where the per-season estimate straddles zero.
cumulative_dirty_air_tax_race_s		Yes		Running race-cumulative sum of per-lap dirty-air tax for each driver, over laps with a measured base. NULL from the first lap with an unknown tax (FD6 / CO-04) onward.
dirtiest_air_lap_in_race_flag		Yes	not_null	TRUE on the lap where this driver’s per-lap tax is at its maximum in the race