Range & domain tests: plausibility, enums, and monotonicity

The three value-shape generic test families, clubbed together because they share one theme: every number and label in the warehouse stays inside the range that’s physically or logically possible.

Three generic dbt/dbt_expectations test kinds, individually small but sharing one job: bounding what a column is allowed to contain. Where Structural asks “does this row exist exactly once,” this group asks “is this value physically or logically possible.”

Bounded shares & probabilities [0, 1]

Anything that’s a fraction, a share, or a probability is range-tested to [0, 1] int_lap_telemetry_aggregates’s throttle/DRS/wheelspin/lift-coast shares (via accepted_range), int_sc_hazard_history’s per-lap SC/VSC/any hazard rates in both raw and shrunk form, and the confidence scores that travel with every shrinkage estimate (affinity_confidence, rating_confidence, tax_calibration_confidence, optimal_pit_lap_confidence, ghost-car recombination_confidence). A value outside [0, 1] here isn’t a plausibility judgment call it’s a unit-error or an off-by-one in a shrinkage formula, full stop.

- name: pct_drs_active
  tests:
    - dbt_utils.accepted_range:
        arguments:
          min_value: 0
          max_value: 1

Honest envelopes

Pace and degradation deltas are bounded to a range wide enough to admit every legitimate outlier (a wet race, a mid-season driver swap) without admitting a decomposition that’s silently blown up. dirty_air_tax_s, the deg-sensitivity slopes (deg_slope_s_per_lap, deg_slope_raw_s_per_lap, and their standard errors), the cliff-onset shift, and both pit-loss estimates (pit_loss_s_empirical, pit_loss_s_shrunk) are each bounded this way wide enough to be honest, narrow enough that a coefficient many multiples outside history fails the build instead of silently propagating into a feature mart.

Enum integrity

accepted_values pins every categorical column to the exact label set the rest of the codebase switches on these sets are themselves gate-checked (a typo’d label fails the build, not a downstream CASE that silently falls through to a default):

Column	Allowed values
compound (multiple models)	SOFT, MEDIUM, HARD, INTERMEDIATE, WET, SUPERSOFT, ULTRASOFT, HYPERSOFT
int_constructor_deg_sensitivity.compound	SOFT, MEDIUM, HARD (slick-only the model that fits degradation slopes)
compound_cliff_params.compound_code (seed)	SOFT, MEDIUM, HARD, INTERMEDIATE, WET
sector (stg_sector_times, int_sector_residual_decomposed)	1, 2, 3
stg_results.dnf_cause	racing, mechanical, non_classified
stg_track_status.status_label	all_clear, yellow, safety_car, vsc, vsc_ending, red_flag, unknown
int_sector_residual_decomposed.dominant_component_class	fuel, compound, rubber, ambient, constructor, dirty_air
int_pit_strategy_value.strategy_verdict	optimal, overran, undercut_forced, early, unknown
int_tyre_surface_vs_bulk_decoupling.degradation_source	surface_driven, bulk_driven, mixed
era / era_key (mart_degradation_history_envelope, int_driver_circuit_era_affinity)	pre2022, post2022

Cross-column monotonicity

Four expect_column_pair_values_A_to_be_greater_than_B tests assert that two related columns stay in the order their meaning requires. int_constructor_structural_pace’s confidence interval must have constructor_structural_pace_ci_high_s ≥ constructor_structural_pace_ci_low_s this is also the test that supersedes the now-placeholder assert_constructor_confidence_monotone singular test, since the model it guarded was folded into this one and re-grained from lap-level to race-level. mart_degradation_history_envelope’s three fuel-removed pace percentiles are pinned to their statistical order: p90 ≥ p50 ≥ p10.

Every model, by range & domain test count

Model	Family	expect_column_values_to_be_between	accepted_range	expect_table_row_count_to_be_between	accepted_values	expect_column_pair_values_A_to_be_greater_than_B	Total
stg_laps	Staging				1		1
stg_results	Staging				1		1
stg_sector_times	Staging				1		1
stg_track_status	Staging				1		1
stg_tyre_allocations	Staging				2		2
dim_circuits	Reference	1					1
dim_corners	Reference				1		1
int_dirty_air_tax_component	Physics	2					2
int_lap_air_state	Physics	1					1
int_lap_normalized_pace	Physics	3					3
int_lap_proximity	Physics	5					5
int_lap_telemetry_aggregates	Physics		4				4
int_lap_thermal_proxy	Physics	1					1
int_stint_geometry	Physics	1					1
int_constructor_structural_pace	Pace Baselines	2				1	3
int_qualifying_segments	Pace Baselines				1		1
int_driver_circuit_affinity	Skill	2					2
int_driver_circuit_era_affinity	Skill	2			1		3
int_driver_season_ratings	Skill	2					2
int_era_normalized_driver_rating	Skill	2					2
int_lap_residual_decomposed	Residual	1					1
int_sector_residual_decomposed	Residual				2		2
int_tyre_surface_vs_bulk_decoupling	Residual	2			1		3
int_constructor_deg_sensitivity	Strategy	6			1		7
int_pit_loss_circuit	Strategy	2					2
int_pit_strategy_cost_curve	Strategy	2			1		3
int_pit_strategy_value	Strategy	4			3		7
int_sc_hazard_history	Strategy	7					7
int_stint_end_regime	Strategy				4		4
fct_cliff_prediction_features	Marts	2					2
fct_driver_skill_features	Marts	1					1
fct_ghost_car_pace	Marts	2					2
fct_ghost_race_finish	Marts	4					4
fct_lap_residuals	Marts				1		1
fct_stint_features	Marts	2			4		6
mart_corner_skill_driver	Marts	4					4
mart_degradation_history_envelope	Marts	3		1	1	3	8
circuit_reference	Reference (seed)	2					2
compound_cliff_params	Reference (seed)	4			1		5
tyre_allocations	Reference (seed)				3		3
Total		72	4	1	31	4	112