Structural tests: grain and completeness across every model

not_null, unique, and unique-combination tests are the largest test family by count and the simplest by idea every model’s declared key is present once and only once.

This is the biggest single page in the CI Contract by test count and the smallest by idea: three dbt test kinds, repeated once per declared key or required column across every staging, reference, intermediate, and mart model (plus the two reference seeds). One pattern, explained once, then a gated table carrying the per-model breakdown.

not_null

The most common test in the suite. Declared on every column a downstream model or the app is allowed to assume is populated a surrogate key, a foreign key, a physics component that’s supposed to exist on every valid lap. A not_null failure means either an upstream join silently dropped rows it shouldn’t have, or a column that’s genuinely optional (and should read NULL downstream, like compound_component_s on a lap with no tyre data) had its test declared too strictly.

- name: lap_id
  description: Foreign key to stg_laps
  tests: [not_null, unique]

unique

Declared on every model’s grain key the column (or surrogate) that should have exactly one row per real-world entity. int_lap_residual_decomposed’s lap_id above is the canonical case: this is the model that closes the seven-term identity, so a duplicated lap_id here wouldn’t just be a data-quality nuisance, it would mean a lap’s driver_skill_residual_s gets counted twice in any downstream aggregate that doesn’t already correct for it.

unique_combination_of_columns

For models whose grain isn’t a single surrogate column, this asserts uniqueness over the column tuple that defines a row. mart_corner_skill_driver is keyed on (race_year, driver_id); mart_degradation_history_envelope is keyed on (circuit_id, era, compound, lap_in_stint) a pre-aggregated cell, not a lap. Two of these exist today, both on marts whose grain is a model-specific aggregate rather than one of the DAG’s two universal grains (lap, sector).

tests:
  - dbt_utils.unique_combination_of_columns:
      arguments:
        combination_of_columns: [circuit_id, era, compound, lap_in_stint]

Why the join spine can't fan out

Every model between stg_laps and a feature mart joins back to the same lap, directly or transitively, by lap_id (or a sector/corner-grain extension of it). The unique test on every model’s grain key is what makes that join spine trustworthy: if any one of the ~30 models a lap passes through on its way to fct_lap_residuals silently duplicated a row, every downstream JOIN on that model would fan out, and a single lap’s contribution to a race-level or season-level aggregate would be counted more than once invisibly, since the duplicated rows would carry consistent-looking values. This is also what protects the identity itself: assert_lap_7term_identity checks that pace_delta_s equals the sum of six components plus the residual on every row of int_lap_residual_decomposed, but that check is only meaningful once per lap. A duplicated lap_id would pass the identity test twice instead of failing it once.

Every model, by structural test count

Model	Family	not_null	unique	unique_combination_of_columns	Total
stg_circuit_info	Staging	4	1		5
stg_events	Staging	1	1		2
stg_laps	Staging	8	1		9
stg_laps_qualifying	Staging	8	1		9
stg_pits	Staging	3	1		4
stg_race_control	Staging	2			2
stg_results	Staging	6	1		7
stg_results_qualifying	Staging	5	1		6
stg_sector_times	Staging	2	1		3
stg_session_status	Staging	3			3
stg_session_status_qualifying	Staging	4			4
stg_track_status	Staging	3	1		4
stg_tyre_allocations	Staging	5			5
dim_circuits	Reference	4	1		5
dim_compounds_season	Reference	4			4
dim_constructors	Reference	2	1		3
dim_corners	Reference	11	1		12
dim_drivers	Reference	1	1		2
int_corner_metrics	Physics	7			7
int_dirty_air_tax_component	Physics	7	1		8
int_lap_air_state	Physics	2	1		3
int_lap_fuel_state	Physics	1	1		2
int_lap_fuel_state_qualifying	Physics	1	1		2
int_lap_normalized_pace	Physics	2	1		3
int_lap_proximity	Physics	3	1		4
int_lap_telemetry_aggregates	Physics	2	1		3
int_lap_thermal_proxy	Physics	2	1		3
int_stint_geometry	Physics	9	1		10
int_track_geometry	Physics	3			3
int_circuit_x_constructor_interaction	Pace Baselines	2	1		3
int_compound_cliff_predicted	Pace Baselines	3	1		4
int_constructor_structural_pace	Pace Baselines	11	1		12
int_constructor_structural_pace_qualifying	Pace Baselines	2	1		3
int_field_pace_curve	Pace Baselines	3			3
int_qualifying_push_laps	Pace Baselines	6	1		7
int_qualifying_segments	Pace Baselines	6	1		7
int_track_evolution	Pace Baselines	3			3
int_constructor_car_fe	Skill	5	1		6
int_driver_circuit_affinity	Skill	12	1		13
int_driver_circuit_era_affinity	Skill	14	1		15
int_driver_race_skill_loro	Skill	5	1		6
int_driver_season_ratings	Skill	8	1		9
int_era_normalized_driver_rating	Skill	12	1		13
int_synthetic_teammate	Skill	4			4
int_corner_skill_residuals	Residual	5	1		6
int_event_corrections	Residual	1	1		2
int_lap_anomaly_flags	Residual	1	1		2
int_lap_corner_inputs	Residual	7	1		8
int_lap_residual_decomposed	Residual	5	1		6
int_lap_residual_decomposed_qualifying	Residual	3	1		4
int_lap_residual_stint_detrend	Residual	2			2
int_qualifying_decomposed	Residual	3	1		4
int_sector_residual_decomposed	Residual	7	1		8
int_tyre_surface_vs_bulk_decoupling	Residual	5	1		6
int_constructor_deg_sensitivity	Strategy	17	1		18
int_pit_loss_circuit	Strategy	5	1		6
int_pit_strategy_cost_curve	Strategy	12		1	13
int_pit_strategy_value	Strategy	4	1		5
int_sc_hazard_history	Strategy	9		1	10
int_stint_end_regime	Strategy	14	1		15
dim_events	Marts	2	1		3
fct_ghost_car_pace	Marts	20	1		21
fct_ghost_race_finish	Marts	9	1		10
fct_lap_residuals	Marts	16	1		17
fct_stint_features	Marts	9	1		10
fct_telemetry_deltas	Marts	9			9
mart_corner_skill_driver	Marts	17		1	18
mart_degradation_history_envelope	Marts	13		1	14
circuit_reference	Reference (seed)	2	1		3
compound_cliff_params	Reference (seed)	7			7
tyre_allocations	Reference (seed)	6			6
Total		421	53	4	478