Table Contracts¶

Every shared table and view in the pipeline, which phase produces it, and which phases consume it. This is the reference for understanding what breaks downstream when you modify a step.

Dependency Map¶

Setup Phase Outputs¶

Object	Type	Produced By	Consumed By
`eaglei_raw`	TABLE	`setup/02`	`setup/03`
`eaglei_part`	TABLE	`setup/03`	`correlation/02`, `correlation/06`
`counties_ref`	TABLE	`setup/04`	`correlation/01`, `correlation/03`, `looker/01`

Correlation Phase Outputs¶

Object	Type	Produced By	Consumed By
`graph_multi_ingredients_hourly`	TABLE	`correlation/01`	`correlation/04`, `correlation/05`, `correlation/07`, `correlation/08`
`view_eaglei_6h_qc`	VIEW	`correlation/02`	`correlation/05`
`view_six_hour_grid`	VIEW	`correlation/03`	`correlation/05`
`view_windhail_thresholds`	VIEW	`correlation/04`	`correlation/07`, `correlation/08`, `correlation/09`
`view_outage_vs_wx_6h_qc`	VIEW	`correlation/05`	`correlation/09`, `correlation/10`, `ml/01`, `looker/01`
`events_restoration`	TABLE	`correlation/06`	`correlation/07`
`event_coverage_wx`	TABLE	`correlation/07`	`looker/01`
`view_daily_plan`	VIEW	`correlation/08`	`looker/01`
`lead_performance`	TABLE	`correlation/09`	(end consumer)
`correlations`	TABLE	`correlation/10`	`looker/01`

ML Phase Outputs¶

Which models are trained is controlled by ML_MODELS in config/.env. Each model produces its own set of output tables with a _{model_key} suffix. The table below shows outputs for all 4 model types.

Shared:

Object	Type	Produced By	Consumed By
`bqml_training_data`	TABLE	`ml/01`	`ml/02a-d`, `ml/03`

Per model (created for each model in ML_MODELS):

Object pattern	Type	Produced By	Notes
`outage_predictor_{key}`	MODEL	`ml/02a-d`	One per model (regressor, classifier, logistic, automl)
`bqml_evaluation_{key}`	TABLE	`ml/03`	Regression metrics (regressor) or classifier metrics (others)
`bqml_feature_importance_{key}`	TABLE	`ml/03`	Feature importance via ML.GLOBAL_EXPLAIN
`bqml_predictions_{key}`	TABLE	`ml/03`	Predicted ratio (regressor) or probability (classifiers) + tier
`bqml_threshold_sweep_{key}`	TABLE	`ml/03`	Regressor only: precision/recall/F1 at cutoffs
`bqml_confusion_matrix_{key}`	TABLE	`ml/03`	Classifiers only: TP/FP/FN/TN
`bqml_roc_curve_{key}`	TABLE	`ml/03`	Classifiers only: ROC curve data

Looker Phase Outputs¶

Object	Type	Produced By	Consumed By
`looker_timeseries_6h`	VIEW	`looker/01`	(end consumer)
`looker_correlation`	VIEW	`looker/01`	(end consumer)
`looker_risk_map`	VIEW	`looker/01`	(end consumer)
`looker_preboard`	VIEW	`looker/01`	(end consumer)
`looker_events`	VIEW	`looker/01`	(end consumer)

Column Schemas¶

graph_multi_ingredients_hourly¶

The primary weather feature table. Most expensive query to produce. All downstream analysis reads from this. Partitioned by DATE(hour_ts), clustered by (county_fips, lead_hours).

Column	Type	Description
county_fips	STRING	5-digit FIPS code (aliased from county_fips_code)
county_name	STRING	County display name
hour_ts	TIMESTAMP	6-hour block start (00/06/12/18 UTC)
lead_hours	INT64	Forecast lead time in hours
ws10_max_mps	FLOAT64	Max 10m wind speed (m/s) across grid cells in county
ws925_max_mps	FLOAT64	Max 925hPa wind speed (m/s)
shear_0_6km_max_mps	FLOAT64	Max 0-6km wind shear (m/s)
updraft700_pos_max_pas	FLOAT64	Max positive 700hPa vertical velocity (Pa/s)
t700_c_min	FLOAT64	Min 700hPa temperature (°C)
t700_c_mean	FLOAT64	Mean 700hPa temperature (°C)
t850_c_mean	FLOAT64	Mean 850hPa temperature (°C)
tp6_mm_max	FLOAT64	Max 6-hour total precipitation (mm)
tp6_mm_mean	FLOAT64	Mean 6-hour total precipitation (mm)

view_outage_vs_wx_6h_qc¶

The master join view. Base for all downstream analysis — weather + outages + QC on the 6-hour grid scaffold.

Column	Type	Description
county_fips	STRING	5-digit FIPS code
valid_ts	TIMESTAMP	6-hour block timestamp (from grid scaffold)
lead_hours	INT64	Forecast lead time in hours
ws10_max_mps	FLOAT64	Max 10m wind speed (m/s)
ws925_max_mps	FLOAT64	Max 925hPa wind speed (m/s)
shear_0_6km_max_mps	FLOAT64	Max 0-6km wind shear (m/s)
updraft700_pos_max_pas	FLOAT64	Max positive 700hPa vertical velocity (Pa/s)
t700_c_min	FLOAT64	Min 700hPa temperature (°C)
t700_c_mean	FLOAT64	Mean 700hPa temperature (°C)
t850_c_mean	FLOAT64	Mean 850hPa temperature (°C)
tp6_mm_mean	FLOAT64	Mean 6-hour total precipitation (mm)
tp6_mm_max	FLOAT64	Max 6-hour total precipitation (mm)
hail_flag	INT64	1 if t700_c_min <= threshold AND tp6_mm_max >= threshold
outage_ratio_6h_max	FLOAT64	Max outage ratio during 6h block
outage_ratio_6h_mean	FLOAT64	Mean outage ratio during 6h block
samples_in_block	INT64	Number of EAGLE-I 15-min samples in the block
customers_out_6h_max	INT64	Max customers out during 6h block

bqml_training_data¶

ML training features + labels derived from the master view.

Column	Type	Description
county_fips	STRING	5-digit FIPS code
valid_ts	TIMESTAMP	6-hour block timestamp
ws10_max_mps	FLOAT64	10m wind speed max (m/s)
ws925_max_mps	FLOAT64	925hPa wind speed max (m/s)
shear_0_6km_max_mps	FLOAT64	0-6km wind shear max (m/s)
updraft700_pos_max_pas	FLOAT64	700hPa updraft max (Pa/s)
tp6_mm_max	FLOAT64	6h precipitation max (mm)
t700_c_min	FLOAT64	700hPa temperature min (°C)
t850_c_mean	FLOAT64	850hPa temperature mean (°C)
hail_flag	INT64	Binary hail indicator
lead_hours	INT64	Forecast lead time in hours
hour_of_day	INT64	Hour of day from valid_ts (0-23)
day_of_week	INT64	Day of week from valid_ts (1=Sun, 7=Sat)
month	INT64	Month from valid_ts (1-12)
wind_precip_interaction	FLOAT64	Derived: ws10_max_mps * tp6_mm_max
wind_squared	FLOAT64	Derived: ws10_max_mps^2
shear_updraft_interaction	FLOAT64	Derived: shear * updraft
outage_ratio_6h_max	FLOAT64	Raw outage ratio (regression label)
outage_event	INT64	Binary label: 1 if outage_ratio >= OUTAGE_THRESHOLD
data_split	STRING	'TRAIN' (80%) or 'TEST' (20%), deterministic by county + date