English for Feature Flag Platform Developers

Feature flag platforms let teams decouple deployment from release, but that flexibility comes with vocabulary a team needs to share precisely — a “rollout,” a “kill switch,” and an “experiment” are three different things with three different risk profiles, and confusing them in a discussion can lead to the wrong safety precautions being taken. This guide covers the English used when discussing feature flag platforms with a team.

Key Vocabulary

Targeting rule — a condition (based on user attributes like plan tier, region, or user ID) that determines which users see which variation of a flag, evaluated on every flag check. “The bug only affected enterprise customers because the targeting rule was scoped to the plan:enterprise attribute — everyone else was on the old code path the whole time.”

Percentage rollout — gradually increasing the share of users who receive a new flag variation, typically to catch problems on a small population before a full release. “Don’t jump straight to 100% — start this at a 5% rollout and watch error rates for an hour before increasing it.”

Kill switch — a flag whose sole purpose is to disable a feature or code path instantly in production without a deploy, used as an operational safety mechanism rather than a release mechanism. “Wrap the new payment retry logic in a kill switch — if it misbehaves under real load, we need to be able to turn it off in seconds, not wait for a rollback deploy.”

Flag debt — the accumulation of old feature flags left in the codebase long after their rollout is complete, adding complexity and risk with no ongoing benefit. “This flag has been at 100% for eight months and every user is on the new path — that’s flag debt, let’s remove the flag and the old code path entirely.”

Stickiness — the guarantee that a given user consistently gets the same flag variation across sessions, usually achieved by hashing a stable user identifier, so users don’t flicker between experiences. “Users were seeing the new checkout flow on one visit and the old one on the next — that’s a stickiness bug, the flag’s bucketing key isn’t stable across sessions.”

A/B test (experiment) flag — a flag used specifically to randomly assign users to variations for statistical comparison, distinct from a simple rollout flag, and usually tied to an analytics or experimentation platform. “This isn’t a rollout flag, it’s an experiment flag — don’t manually override anyone’s assignment, or you’ll bias the statistical results we’re trying to measure.”

Common Phrases

• “Is this a kill switch, a rollout flag, or an experiment flag — because the safety expectations are different for each?”
• “What’s the current rollout percentage, and what’s the plan for increasing it?”
• “Is this flag still needed, or is it flag debt we should clean up?”
• “Is this user’s assignment sticky across sessions, or could it flicker?”
• “Is anyone manually overriding assignment on this experiment flag?”

Example Sentences

Reviewing a pull request: “This flag has been fully rolled out for months with no plan to remove it — let’s add a ticket to delete the flag and the fallback code path, this is flag debt.”

Explaining a design decision: “We wrapped the new recommendation algorithm in a kill switch separate from its rollout flag, so ops can disable it instantly during an incident without touching the rollout percentage.”

Describing a bug: “Users kept flip-flopping between the old and new UI because the flag was bucketing on session ID instead of user ID — switching to a stable user-based key fixed the stickiness issue.”

Professional Tips

• Say “kill switch” specifically for an emergency-disable flag — using “flag” generically for both rollout and kill-switch purposes muddies the operational risk conversation.
• When reviewing flag usage, ask “is this flag debt?” — a flag stuck at 100% or 0% with no active purpose should trigger a cleanup ticket, not be left indefinitely.
• Use “stickiness” precisely when describing consistent per-user assignment — it’s the standard term and distinguishes the concern from targeting rules or rollout percentage.
• Distinguish an “experiment flag” (statistical assignment, don’t manually override) from a “rollout flag” (progressive release, manual overrides for support cases are often fine) when setting expectations with support teams.

Practice Exercise

1. Explain in two sentences the difference between a kill switch and a percentage rollout.
2. Write a one-sentence code review comment flagging an old flag as flag debt.
3. Describe, in your own words, why manually overriding a user’s assignment on an experiment flag is a problem.