Skip to main content

Data Dictionary

Complete reference of data fields and types used across Snowtrail products.

Point-in-Time Columns

Every Snowtrail table, signals, features, events, and backtests, includes two point-in-time columns that support reproducible research and audit.

ColumnTypeDescription
str_effective_atdateThe date this row became knowable: the business date it describes, plus the source's publication lag where one applies. This is the column the as_of parameter filters on.
str_ingested_attimestampThe UTC timestamp of the pipeline run that last wrote this row. Rewritten on every rebuild, not a first-publication date, and not used by as_of.

as_of filters on str_effective_at, so a query replayed at a historical date returns only what the market could actually have seen by then. This eliminates look-ahead bias arising from publication lag. It does not reconstruct values that a source has since restated, see Point-in-Time Integrity for what as_of does and does not do.

Common API Response Fields

When querying the Snowtrail API, responses include standard envelope fields.

FieldTypeDescription
product_idstringIdentifier for the product (e.g. gbsi_us, glmi)
countintegerNumber of records returned in this response
has_morebooleanWhether additional pages of results exist
next_cursorstringCursor token for retrieving the next page of results

Signal Columns

Signal schemas are product and dataset-specific. Below is an example using GBSI-US system stress.

Example: GBSI-US System Stress

ColumnTypeDescription
week_endingdateThe week-ending date for this observation
stress_regimeinteger (1--5)Numeric stress regime classification, where 1 is lowest stress and 5 is highest
stress_regime_labelstringHuman-readable label for the stress regime
confidence_scorefloatConfidence in the regime classification
directional_biasstringDirectional context associated with the current regime
str_effective_atdateBusiness date for this row
str_ingested_attimestampWhen this row was processed by Snowtrail

Other signal datasets (e.g. balance momentum, price context, grid stress) follow similar patterns but with columns specific to their domain.

Feature Columns

Feature schemas vary by product and dataset. Features represent the derived, research-ready inputs that underpin signals and events. Every feature table includes:

ColumnTypeDescription
str_effective_atdateBusiness date for this row
str_ingested_attimestampWhen this row was processed by Snowtrail

Additional columns are specific to the product and feature dataset. For example, GBSI-US storage features include columns for storage levels, injections, withdrawals, and year-over-year comparisons.

Event Columns

Event schemas vary by product and dataset. Events represent discrete developments that materially affect market conditions. Every event table includes:

ColumnTypeDescription
str_effective_atdateBusiness date for this row
str_ingested_attimestampWhen this row was processed by Snowtrail

Additional columns describe the event type, severity, status, and other product-specific attributes.

Schema Discovery

Actual column names and types are product and dataset-specific. To explore the schema for a particular dataset:

  • Use the Python SDK to fetch a sample and inspect the resulting DataFrame columns
  • Consult the API Reference for endpoint-specific documentation
  • Contact support@snowtrail.ai for detailed schema documentation