int_stint_end_regime

Why did this stint end?

Part of the Strategy family.

Why did this stint end? One row per stint. The stint-life survival target treats every uncensored stint as a completed tyre life, but 28.8% of them end under a safety car, VSC or red flag - the moment the pit wall was handed a free stop, not the moment the tyre ran out. This model names the cause so a consumer can separate the tyre-limit process from the interruption process instead of fitting their mixture. It also owns is_censored_stint, which fct_stint_features reads from here, so “the driver’s last stint of the race” has one definition. Precedence is declared in the model header: censoring outranks regime, and within an uncensored stint red_flag > safety_car > vsc > green.

Lineage

Overview

Property	Value
Layer	intermediate
Family	Strategy
Contract enforced	No
Upstream	int_stint_geometry, stg_results

20 tests guard this model - 19 generic, 1 singular.

Columns

Column	Type	Nullable	Tests	Description
stint_id		Yes	not_null, unique	PK from int_stint_geometry. One row per stint.
race_year		Yes	not_null	Season year.
race_id		Yes	not_null	Race identifier (numeric FastF1 event id cast to string).
driver_id		Yes	not_null	FastF1 three-letter driver code (e.g. VER, HAM).
stint_number		Yes		Stint ordinal within the driver’s race. NULL for the 325 stints in 2018 built from laps FastF1 never assigned to a stint; those carry stint_end_cause NULL and end_cause_confidence ‘unassigned_stint’.
end_lap_id		Yes	not_null	Foreign key to stg_laps - the lap the stint ended on, chronologically. For an uncensored stint this is the in-lap.
end_lap_number		Yes	not_null	Race lap number of the stint’s final lap.
end_lap_in_stint		Yes	not_null	Chronological ordinal of the stint’s final lap within the stint (equals stint_length_actual).
is_censored_stint		Yes	not_null, accepted_values	TRUE when this is the driver’s last stint of the race, so it ended at the flag or at retirement rather than at a tyre change. Remaining stint life on these laps is a lower bound, not an observation. Consumed by the stint-life survival model (ml/src/features.py) to set the AFT upper bound to +inf, and by the blind-test scoreboard to report censored and uncensored stints apart. Never a model feature.
end_regime		Yes	not_null, accepted_values	Track regime in force on the stint’s final lap, for every stint including censored ones. Severity precedence red_flag > safety_car > vsc > green, because one TrackStatus string can carry several codes at once. Crossed with is_censored_stint this reproduces the measured end-regime decomposition.
is_multi_regime_end		Yes	not_null	TRUE when the final lap carried more than one of red/SC/VSC, so severity precedence had to choose which one names the ending.
ended_on_pit_lap		Yes	not_null	TRUE when the stint’s final lap carries a pit marker. FALSE on an uncensored stint means the tyre change is known from the existence of a later stint but is not visible on the lap itself.
was_running_at_flag		Yes	not_null	TRUE when the driver’s classified status says the car was still circulating at the chequered flag (‘Finished’, ‘+N Lap(s)’, ‘Lapped’). Disqualification is resolved on distance instead, because it is a post-race ruling rather than a reason a stint ended. Only meaningful on a censored stint.
stint_end_cause		Yes	not_null, accepted_values	Six-class answer to “why did this stint end”: green_pit, sc_pit, vsc_pit, red for the four uncensored causes, race_end and retirement for the two censored ones. NULL where stint_number is NULL - the row is an artefact of unassigned laps, not a stint that ended.
end_cause_confidence		Yes	not_null, accepted_values	How confidently the cause is known. ‘observed’ where every input was read; ‘inferred’ where severity precedence had to choose, where the pit marker is missing on an uncensored stint, or where disqualification hides whether the driver was running; ‘unassigned_stint’ where the row is not a stint.