fct_ghost_car_pace

One row per (ego_driver, host_constructor, race, lap).

Part of the Marts family.

One row per (ego_driver, host_constructor, race, lap). “What if” lap times: recombines ego driver’s skill (residual) with host constructor’s car pace. predicted_lap_time_s = ego skill + host car. Degenerate check: ego == host → predicted == actual. Powers “ghost car” simulator: replay with different cars.

Lineage

Overview

Property	Value
Layer	marts
Family	Marts
Contract enforced	No
Tags	simulation, ghost_car
Upstream	int_lap_residual_decomposed, int_compound_cliff_predicted, stg_laps, int_constructor_structural_pace, int_circuit_x_constructor_interaction, int_constructor_deg_sensitivity

25 tests guard this model - 23 generic, 2 singular.

Columns

Column	Type	Nullable	Tests	Description
ghost_id		Yes	not_null, unique	Surrogate PK = MD5(ego_driver_id || host_constructor_id || race_id || lap_number)
predicted_lap_time_s		Yes	not_null	Recombined lap time using ego skill + host constructor pace.
actual_lap_time_s		Yes	not_null	Observed actual lap time for the ego driver.
predicted_residual_pace_s		Yes	not_null	predicted_lap_time_s with the fuel-burn component backed out (the fuel term is simply omitted from the recombination sum). Fuel-adjusted so ranking by pace isn’t biased by which lap window (heavy vs. light fuel) a driver ran.
actual_residual_pace_s		Yes	not_null	actual_lap_time_s minus fuel_component_s. Fuel-adjusted counterpart to actual_lap_time_s.
delta_vs_actual_lap_s		Yes	not_null	predicted_lap_time_s minus actual_lap_time_s. 0 when ego == host.
deg_interaction_s		Yes	not_null	(host deg slope - ego deg slope) x age_in_stint, slopes from int_constructor_deg_sensitivity at the ego lap’s (compound, season). The only host-dependent term that varies within a race, so it is what lets different hosts produce different driver orders. Exactly 0 when ego == host; 0 when no cell exists (wet compounds / low-sample = field-average degradation).
host_deg_slope_s_per_lap		Yes	not_null	Host constructor’s deg slope for the ego lap’s compound; 0 when no cell.
ego_deg_slope_s_per_lap		Yes	not_null	Ego constructor’s deg slope for the ego lap’s compound; 0 when no cell.
cliff_interaction_s		Yes	not_null, expect_column_values_to_be_between	severity x [hinge(host) - hinge(ego)] where hinge(c) = GREATEST(0, age - field_onset - cliff_onset_shift(c)). The nonlinear host-dependent term: it only fires once tyres approach/pass the cliff, reordering drivers on long stints. Exactly 0 when ego == host. Guardrail-clipped to +/-2 s.
host_cliff_shift_laps		Yes	not_null	Host constructor’s cliff-onset shift (laps) for the ego lap’s compound; 0 when no cell.
ego_cliff_shift_laps		Yes	not_null	Ego constructor’s cliff-onset shift (laps) for the ego lap’s compound; 0 when no cell.
recombination_confidence		Yes	not_null, expect_column_values_to_be_between	Continuous confidence in recombination [0, 1] = host_obs_factor (panel_obs / (panel_obs + 50)) × the lap’s correction_weight. Degrades smoothly when the host has sparse panel data or the lap was soft-down-weighted.
host_constructor_pace_se_s		Yes	not_null	SE-propagation ingredient: SE of the host constructor’s structural pace (0 when no panel cell). Aggregated into predicted_mean_lap_se_s downstream; cancels in within-scenario ordering since every driver shares the host car.
host_deg_slope_sd_s_per_lap		Yes	not_null	Posterior sd of the host deg slope for the ego lap’s compound; 0 when no cell.
ego_deg_slope_sd_s_per_lap		Yes	not_null	Posterior sd of the ego deg slope for the ego lap’s compound; 0 when no cell.
host_cliff_shift_se_laps		Yes	not_null	SE (laps) of the host cliff-onset shift; 0 when no cell.
ego_cliff_shift_se_laps		Yes	not_null	SE (laps) of the ego cliff-onset shift; 0 when no cell.
host_cliff_active		Yes	not_null	1.0 when the host car is past its shifted cliff onset on this lap, else 0.0.
ego_cliff_active		Yes	not_null	1.0 when the ego car is past its shifted cliff onset on this lap, else 0.0.