Skip to main content

Features Overview

Features are curated, transformed datasets created from raw fundamental inputs, designed as reliable building blocks for quantitative research.

What is a Feature?

A Feature is a derived dataset that applies consistent definitions and transformations to raw data, producing stable, research-ready inputs. Features are designed to:

  • Reduce data preparation overhead
  • Ensure consistent definitions across analyses
  • Maintain point-in-time integrity for backtesting

Feature Structure

Each feature includes:

Column names vary by product and pillar. Taking gbsi_us.fea_storage_inventory as the example:

FieldDescription
effective_dateThe business date the row describes
storage_bcfThe measured value for that date
storage_5y_avgFive-year average for the same period
storage_vs_5y_pctDeviation from that average, as a percentage
storage_zscoreStandardised deviation, comparable across series
str_effective_atWhen the value became publicly knowable
str_observed_atWhen Snowtrail could first have held it
str_ingested_atWhen this row was written

Every feature table carries the three PIT timestamps. The domain columns differ by pillar; see each product's Features page for the full list.

Feature Categories

Features are organised by data domain:

  • Supply features - Production, imports, capacity
  • Demand features - Consumption, exports, usage patterns
  • Storage features - Inventory levels, injection/withdrawal
  • Infrastructure features - Capacity, utilisation, constraints
  • Weather features - Temperature, degree days, anomalies