Skip to main content

Feature Design Principles

Snowtrail Features are built with research integrity as a priority.

Core Principles​

1. Point-in-Time Construction​

Features reflect what was knowable at the time. Every row carries the date its inputs became available, so a feature is never surfaced earlier than the source data behind it. Values reflect each source's current best figure; where a source later restates, the feature is recomputed on the next scheduled run.

2. Stable Definitions​

Feature logic is consistent and documented. Every row carries a model_version, so you can tell whether a series you hold was produced under the same definition as one you pull today. See Versioning.

3. Cross-Market Consistency​

Common frameworks are applied across markets where appropriate, enabling meaningful comparisons and combined analyses.

4. Research-First Design​

Features are built for modelling and analysis, not just visualisation. This means attention to:

  • Appropriate granularity
  • Handling of missing values
  • Alignment of time series
  • Clear unit definitions

Transformation Standards​

Features apply consistent transformations:

  • Temporal alignment - Data aligned to consistent time boundaries
  • Unit standardisation - Consistent units within feature families
  • Missing value handling - Documented approach to gaps
  • Revision handling - Source restatements flow through on the next scheduled run

Documentation​

Each feature includes:

  • Definition and methodology
  • Data sources
  • Update frequency
  • Known limitations