Template

Feature Registry: Minimum Viable Template

An interactive, team-ready feature registry entry form that captures canonical definitions, provenance, transformation links, validation and monitoring notes, ownership, freshness, privacy sensitivity, and onboarding steps — making features discoverable, reusable, and governable.

Interactive Tool

Feature Registry Entry

This form captures a minimum viable, operational feature record suitable for discovery, reuse, governance, and handoff between analytics and engineering teams. Use one submission per distinct feature (canonical name + definition). Provide examples, links to code or repo, and practical notes that reduce rework, prevent leakage, and speed productionization.

The registry entry emphasizes reproducibility (canonical SQL or transformation), ownership, freshness, validation evidence, monitoring signals, and special risks such as privacy or distributional bias.

A short canonical name used across pipelines and docs (snake_case recommended). E.g., user_purchase_rate
Two or three sentences describing what the feature measures and why it's useful.
Representative SQL, pseudocode, or formula that clearly defines how the feature is computed. Include any key joins, filters, windowing, or aggregations so others can reproduce it.
Link to the canonical transformation, feature pipeline, notebook, or feature store entry (git, artifact, or feature-store URI).
List the authoritative source tables, streams, APIs, or external datasets with schema versions if applicable.
List the specific columns or fields used and any column-level transformations (e.g., normalization, encoding).
How often this feature is refreshed or recomputed in production.
Team, role, or person responsible for the feature's correctness, freshness, and monitoring (include preferred contact/email).
Semantic version, commit hash, or date-based versioning for the canonical definition.
Date when this feature's definition and sample distribution were last reviewed (ISO YYYY-MM-DD).
Short summary of observed distribution (mean, median, skew, common ranges) and any known edge cases. Attach or link to a sample dataset if available.
Numeric minimum typically observed (if applicable). Useful for automated sanity checks.
Numeric maximum typically observed (if applicable).
Helps engineers choose storage and materialization strategies.
Conditions or filters used to include the feature in training datasets (e.g., time windows, population filters). Note how to avoid label leakage.
Conditions or logic applied when computing or using the feature at inference/serving time.
Document known limitations — demographic, geographic, device, or business-segment gaps — and recommended mitigations.
Quick flag that indicates whether this feature is likely to shift over time and should be monitored closely.
Indicate whether PII or regulated data is involved and what handling constraints apply.
Yes if the feature directly or indirectly includes personally identifying information.
List models, dashboards, decision systems, or reports that rely on this feature (include links if possible).
E.g., freshness lag, null rate, distribution drift tests, cardinality changes, sudden spikes. Include alert thresholds if known.
Brief summary of validation results that support the feature's usefulness (e.g., feature importance, AUC lift, ablation study notes). Link to notebooks or experiments.
Notes about derivation, parent features, or related features that may be duplicates or alternatives.
Comma-separated tags to aid discovery (e.g., user, engagement, finance, device).
Link to a small sample data file, dashboard, or storage path used for QA or review.
Practical steps for safely adding this feature to training or production: e.g., test queries, unit tests, monitoring setup, rollout plan, rollback criteria.
Describe potential leakage pathways (look-ahead bias, derived from label) and concrete mitigations used in experiments and production.
Any other practical guidance for users or engineers working with this feature.
You can explore this tool now. Sign in or create an account to save your responses and return to them later.
Sign In to Save Create an Account
Make this tool part of your work

Save a personal copy, bring it to your team, or tailor the questions and workflow to fit what you are hungry to improve.

Member customization and team collaboration are coming soon.
Discuss Ask the Librarian

Discussion

Comments and conversation will live here.