fct_ghost_race_finish

One row per (host_constructor, ego_driver, race).

Part of the Marts family.

One row per (host_constructor, ego_driver, race). Projects finishing position if ego driver raced in host constructor’s car, ranked by predicted MEAN lap pace (invariant to lap count, so DNF / partial runs are ranked fairly and flagged via is_short_run). Filtered to recombination_confidence ≥ 0.3. Powers “what if” race outcome scenarios.

Lineage

Overview

Property	Value
Layer	marts
Family	Marts
Contract enforced	No
Tags	simulation, ghost_car
Upstream	fct_ghost_car_pace, stg_laps, stg_results

16 tests guard this model - 14 generic, 2 singular.

Columns

Column	Type	Nullable	Tests	Description
ghost_race_id		Yes	not_null, unique	Surrogate PK = MD5(race_year || race_id || ego_driver_id || host_constructor_id)
is_self_scenario		Yes	not_null	True when ego driver is placed in their own constructor’s car (degenerate identity).
predicted_finish_position		Yes	not_null, expect_column_values_to_be_between	Projected finish position within this host-constructor scenario (ranked by mean pace).
actual_finish_position		Yes		Actual finishing position, sourced from official classified results (stg_results) classified finishers get their official position, retirements get NULL. Falls back to the legacy last-valid-lap position only for races not yet in stg_results (see finish_from_results).
actual_is_dnf		Yes		True when official results classify the driver as a retirement (NULL position). FALSE when results absent.
actual_dnf_cause		Yes		Coarse DNF cause from stg_results (‘racing’/‘mechanical’/‘non_classified’); NULL for finishers or races without results.
finish_from_results		Yes		True when actual_finish_position came from official classified results; False when it fell back to the lap-derived position.
delta_vs_actual_position		Yes		Projected minus actual finish position. Negative = would have finished higher. NULL when the driver has no real finishing position (DNF).
predicted_mean_lap_s		Yes	not_null	Mean predicted lap time across counted laps; the ranking key.
predicted_mean_residual_pace_s		Yes	not_null	Mean fuel-adjusted predicted pace (AVG of fct_ghost_car_pace’s predicted_residual_pace_s over counted laps). The ranking key for assert_ghost_self_scenario_rank so drivers aren’t ranked on raw pace that conflates fuel load with true speed.
actual_mean_residual_pace_s		Yes	not_null	Mean fuel-adjusted actual pace (AVG of actual_residual_pace_s over counted laps).
lap_coverage		Yes		laps_counted / race_distance_laps in [0,1]; fraction of the race the driver contributes.
is_short_run		Yes	not_null	True when lap_coverage < ghost_short_run_threshold (default 0.5) DNF / partial run.
avg_recombination_confidence		Yes		Coverage-weighted lap confidence (n/(n+50) heuristic × sqrt coverage), used by the Counterfactual Championship and Hidden Performance app features. Coexists with the finer-grained SE columns below rather than being replaced by them.
predicted_mean_lap_se_s		Yes	not_null, expect_column_values_to_be_between	Marginal standard error (seconds) of predicted_mean_lap_s, propagating host structural-pace SE, host/ego deg-slope posterior sds, host/ego cliff-shift SEs (scaled by tyre-age and past-cliff exposure), and the finite-sample SE of the lap mean. Includes the host-pace term that cancels in within-scenario ordering.
p_beats_next		Yes	expect_column_values_to_be_between	Probability this driver finishes ahead of the driver ranked immediately below it in this scenario, under a normal approximation on the pairwise pace gap (host structural pace cancels). NULL for the last-ranked driver (no one below). ≥ 0.5 by construction since the driver above always has the lower predicted pace.
finish_pos_se		Yes	not_null, expect_column_values_to_be_between	Standard error of the projected finish position the Poisson-binomial sd of the number of drivers predicted ahead (sqrt of sum p_k(1-p_k) over pairwise order probabilities). 0 when the scenario has a single driver.