Feature Flags & Rollouts

This guide walks you through auditing a project's feature flag system - gradual rollouts, A/B testing capabilities, kill switches, and targeting.

The Goal: Deployments Decoupled from Releases

Ship code anytime, release features when ready. Feature flags give you control over what users see without redeploying.

Enabled — A feature flag system exists and is actively used (env vars for small projects, dedicated service for larger ones)
Gradual — Percentage-based rollouts, sticky user assignment, and segment targeting are available
Kill switches — Critical features (payments, external APIs, new features) can be toggled without a deploy
Instant — Kill switch toggle time is under 5 minutes, ideally instant via dashboard
Documented — Runbooks exist for disabling features during incidents

Before You Start

Identify project scale (small projects can use env vars, larger projects need dedicated service)
Identify feature flag service (GrowthBook, LaunchDarkly, Unleash, Flagsmith, custom)
Identify deployment frequency (frequent deploys benefit more from feature flags)
Check for existing flags (are they actually in use or just installed?)

general

FF-001

Feature flag system configured recommended

Feature flag system exists (env vars for small projects, GrowthBook for larger) with capabilities for percentage rollouts, A/B testing, sticky assignment, and user segment targeting

“How do you ship to 5% of users without a deploy?”

FF-002

Kill switches for quick disable recommended

Critical features have kill switches that can be toggled in < 5 minutes without deploy, with documented procedures and controlled access

“How fast can you disable that new feature if it's melting prod?”