int_constructor_deg_sensitivity

One row per (race_year, constructor_id, compound).

Part of the Strategy family.

One row per (race_year, constructor_id, compound). Constructor-specific tyre degradation sensitivity: within-stint fixed-effects slope of the post-decomposition residual on tyre age, field-centred per (compound, season) and empirical-Bayes shrunk toward 0 (DerSimonian-Laird). Positive = degrades faster than the field-average compound curve. Feeds the ghost-car recombination interaction term (host slope minus ego slope) x age_in_stint, which is what lets host identity reorder drivers. slope_se and n_laps are carried for downstream SE propagation. cliff_onset_shift_laps is a per-constructor break-point offset (laps) for the field tyre-cliff onset, estimated as a bivariate within-stint hinge coefficient (extra post-onset degradation), field-centred and EB-shrunk like the linear slope, then mapped to an onset shift via the field severity and typical post-onset depth. Positive = cliff arrives later (gentler). Feeds the ghost-car cliff interaction term.

Lineage

Overview

Property	Value
Layer	intermediate
Family	Strategy
Contract enforced	No
Tags	causal_decomposition, constructor, degradation
Upstream	int_lap_residual_decomposed, int_compound_cliff_predicted, dim_compounds_season

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

Columns

Column	Type	Nullable	Tests	Description
deg_sensitivity_id		Yes	not_null, unique	Surrogate PK = race_year || constructor_id || compound
race_year		Yes	not_null	Season year (cells are season-scoped; no cross-season pooling)
constructor_id		Yes	not_null	Constructor identifier
compound		Yes	not_null, accepted_values	Dry tyre compound the slope is estimated for
deg_slope_s_per_lap		Yes	not_null, expect_column_values_to_be_between	Production value: EB-shrunk deviation from the field-mean within-stint slope, in s/lap per lap of tyre age. Positive = degrades faster than field. 0.0 for low-sample cells. 2024 reference range: ~±0.025 s/lap on HARD, ~±0.08 on SOFT.
deg_slope_raw_s_per_lap		Yes	expect_column_values_to_be_between	Uncentred within-stint FE slope (diagnostic). Uniformly negative (~ -0.08 s/lap in 2024) because a common residual fuel under-correction survives the decomposition; the common part cancels in field-centring and in the (host - ego) recombination difference.
field_mean_slope_s_per_lap		Yes		Precision-weighted mean raw slope per (race_year, compound); the centring constant.
deg_slope_se_s_per_lap		Yes	expect_column_values_to_be_between	FE-regression standard error of the raw slope (dof = n_laps - n_stints - 1), floored at 1e-4. NULL for low-sample cells. Carried for downstream SE propagation.
deg_slope_posterior_sd_s_per_lap		Yes	not_null	EB posterior sd of the shrunk slope; equals tau (prior sd) for low-sample cells and 0 when tau^2 = 0 (no between-constructor signal).
shrink_factor		Yes	not_null, expect_column_values_to_be_between	tau^2 / (tau^2 + se^2) in [0,1]; 0 for low-sample cells.
tau_s_per_lap		Yes	not_null	DerSimonian-Laird between-constructor sd per (race_year, compound).
n_laps		Yes	not_null	Clean dry pre-cliff laps in the cell (cell-count; qualifying rule >= 30).
n_stints		Yes	not_null	Distinct stints in the cell (qualifying rule >= 5).
is_low_sample		Yes	not_null	TRUE when the cell fails minimum-cell rules; slope forced to 0, excluded from pooling.
cliff_onset_shift_laps		Yes	not_null, expect_column_values_to_be_between	Per-constructor tyre-cliff onset shift in laps relative to the field onset. Positive = cliff arrives LATER (gentler), negative = EARLIER (harsher). 0.0 for cells without enough post-onset depth. Clipped to +/-3 laps.
cliff_onset_shift_se_laps		Yes	expect_column_values_to_be_between	SE of the onset shift (hinge SE mapped through ref_depth / severity). NULL for low-sample cells. Carried for downstream SE propagation.
cliff_hinge_coef_s_per_lap		Yes		Diagnostic: raw bivariate within-stint hinge coefficient (extra s/lap of degradation after the field onset). Uniformly negative like the linear slope; common part cancels in field-centring.
cliff_hinge_field_mean_s_per_lap		Yes		Precision-weighted mean hinge coefficient per (race_year, compound); the centring constant.
cliff_severity_used_s_per_lap		Yes		Median field post-onset severity (s/lap) per (compound, season), floored at 0.30, used in the shift mapping.
cliff_ref_depth_laps		Yes		Mean laps past the field onset over post-onset clean laps per (compound, season); the depth at which the hinge deviation is converted to an onset shift.
cliff_hinge_tau_s_per_lap		Yes		DerSimonian-Laird between-constructor sd of the hinge coefficient per (race_year, compound).
n_post_cliff_laps		Yes	not_null	Post-onset clean laps in the cell (qualifying rule >= 30).
n_deep_cliff_laps		Yes	not_null	Post-onset laps at least 2 laps past onset (depth rule >= 15); guards against shallow-crossing noise.
n_stints_cliff		Yes	not_null	Distinct stints contributing post-onset laps (qualifying rule >= 5).
is_low_sample_cliff		Yes	not_null	TRUE when the cell fails cliff minimum-cell rules; cliff_onset_shift_laps forced to 0.
fit_timestamp		Yes	not_null	UTC timestamp when the slopes were fitted (ISO 8601)