int_pit_strategy_cost_curve

One row per (stint_id, horizon_scope, candidate_pit_lap_offset): the Total_Cost(L) surface int_pit_strategy_value’s header described for two years and never computed.

Part of the Strategy family.

One row per (stint_id, horizon_scope, candidate_pit_lap_offset): the Total_Cost(L) surface int_pit_strategy_value’s header described for two years and never computed. Cost of pitting after L laps on the current set = wear on the old set over those laps + wear on the next set over the rest of the horizon + a compound baseline offset + pit-lane loss discounted by the chance a safety car has already appeared. L = H is the no-stop candidate, which pays no pit term and is what makes a stop have to justify itself. THE SC DISCOUNT IS LOAD-BEARING: under a fixed one-stop the pit loss is the same whenever you take it, so it adds a constant to every candidate and cancels out of the argmin entirely - a longer pit lane would move nothing. Discounting it by 1 - (1 - m)(1 - (1 - h)^L), with h from int_sc_hazard_history, is what makes waiting pay and gives the model the property assert_pit_loss_pushes_optimum_later checks. Per-lap wear is capped at var(‘compound_wear_max_s_per_lap’) — the shared bound, which int_compound_cliff_predicted now applies to the same three terms as of Phase 8. The cliff polynomial is unbounded, reaching 136 s/lap extrapolated to age 80 and already 93 s/lap in-sample.

Lineage

Overview

Property	Value
Layer	intermediate
Family	Strategy
Contract enforced	No
Tags	simulation, strategy
Upstream	int_stint_geometry, stg_weather, dim_compounds_season, int_pit_loss_circuit, int_sc_hazard_history
Model-level tests	unique_combination_of_columns

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

Columns

Column	Type	Nullable	Tests	Description
stint_id		Yes	not_null	FK to int_stint_geometry. Part of the compound PK.
horizon_scope		Yes	not_null, accepted_values	window = this stint’s first lap through the next stint’s last; race = through the driver’s last lap of the race. Both are emitted for every stint that has them.
horizon_laps		Yes	not_null	Length of the horizon in valid laps.
candidate_pit_lap_offset		Yes	not_null	Laps run on the current set before boxing, in valid laps. Runs 1..horizon_laps; the last value is the no-stop candidate. Part of the compound PK.
laps_on_new_set		Yes	not_null	horizon_laps minus the offset. Zero for the no-stop candidate.
old_wear_cost_s		Yes	not_null	Cumulative wear over the laps run on the current set, evaluated on its own fitted curve from the age the stint opened on - a stint started on a used set begins part-way up that curve rather than at zero.
new_wear_cost_s		Yes	not_null	Cumulative wear over the rest of the horizon on the next set, from age 1.
baseline_cost_s		Yes	not_null	Per-lap pace offset between the two compounds (compound_grip_peak plus the optimal-temperature term) over the laps on the new set. Both inputs are per-compound CONSTANTS in compound_cliff_params, never fitted, and grip_peak as typed prices SOFT slower than HARD. var(‘pit_strategy_baseline_delta’) decides whether this term is allowed to move the argmin; setting it FALSE zeroes this column.
sc_hazard_per_lap		Yes	not_null	Per-racing-lap SC/VSC hazard for the circuit, EB-shrunk, from int_sc_hazard_history. 0.0 for a circuit with no ingested track_status timeline, which makes the discount constant and returns the model to the pit-loss-cancels case.
sc_probability		Yes	expect_column_values_to_be_between	P(a caution has appeared within the candidate’s laps) = 1 - (1 - h)^L.
pit_discount_factor		Yes	not_null, expect_column_values_to_be_between	Fraction of the green-flag pit-lane loss this candidate expects to pay. Non-increasing in the candidate lap - asserted by assert_pit_discount_monotone - and a hard 0 for the no-stop candidate.
wear_cost_s		Yes	not_null	Everything that does not scale with pit_lane_loss_s. Split out so a consumer - or assert_pit_loss_pushes_optimum_later - can re-minimise at a different pit loss without rebuilding the curve.
total_cost_s		Yes	not_null	wear_cost_s + pit_lane_loss_s x pit_discount_factor. The argmin target.