int_constructor_structural_pace

One row per (race_year, race_id, constructor_id).

Part of the Pace Baselines family.

One row per (race_year, race_id, constructor_id). Car-inherent pace advantage/penalty estimated from a re-centred median pace delta across same-constructor teammates in the same race, isolating car pace from driver skill via within-constructor, between-driver variation. A placeholder for the full panel regression (pyfixest HDFE fit) named in the model’s own header; that fixed-effects estimate exists separately, as int_constructor_car_fe. Positive = slower car. Includes a normal- approximation 95% CI and clean pair sample counts for confidence.

​

Lineage

​

Overview

Property	Value
Layer	intermediate
Family	Pace Baselines
Contract enforced	No
Tags	causal_decomposition, constructor
Upstream	int_lap_fuel_state, int_field_pace_curve, int_stint_geometry, stg_laps, int_event_corrections, int_track_evolution
Model-level tests	expect_column_pair_values_A_to_be_greater_than_B

18 tests guard this model - 17 generic, 1 singular.

​

Columns

Column	Type	Nullable	Tests	Description
constructor_race_id		Yes	not_null, unique	Surrogate PK = hash(race_year, race_id, constructor_id)
race_year		Yes	not_null	Season year
race_id		Yes	not_null, relationships	Race identifier
constructor_id		Yes	not_null, relationships	Constructor identifier (FK to dim_constructors)
constructor_structural_pace_s		Yes	not_null	Re-centred median pace-delta coefficient in seconds (not yet a fitted fixed effect see the model description). Positive = slower than field. Identified via within-constructor, between-driver variation from teammate pairs. NULL (unknown, not 0.0) when 10 or fewer of the team’s clean laps have a measured field base (CO-04): every team of 2018_1 / 2018_2 (no telemetry, FD6) and of 2018_3 / 2025_6 (tyre data quarantined by stg_lap_tyre_qa, so no lap has a tyre cost or a base), which used to read 0.000 through a COALESCE(base, lap_time_s).
constructor_structural_pace_se_s		Yes	not_null, expect_column_values_to_be_between	Standard error of the median pace delta (stddev / √n), not yet a clustered (CRV1) regression SE. Positive, ≥0. Captures uncertainty in the coefficient estimate. NULL where the coefficient is.
constructor_structural_pace_ci_low_s		Yes	not_null	95% confidence interval lower bound. NULL where the coefficient is.
constructor_structural_pace_ci_high_s		Yes	not_null	95% confidence interval upper bound. NULL where the coefficient is.
panel_observations_n		Yes	not_null	Clean laps with a measured field base that the estimate rests on (CO-04: laps with no base are not counted). A row exists for every (race, constructor) with more than 10 clean laps; the coefficient needs more than 10 of them measured.
clean_teammate_pair_laps_n		Yes	not_null	Count of clean laps where both teammates in a constructor ran (used to assess pair quality). Flags constructor-seasons where this is <100 laps.
r_squared_within		Yes	expect_column_values_to_be_between	Within fixed-effect R² (sanity check; low is expected, high is good). NULL until the pyfixest model is in place a misleading 0.0 is worse than NULL.
fit_timestamp		Yes	not_null	UTC timestamp when this coefficient was fitted (ISO 8601)