name: Split Vocabulary description: >- Normative vocabulary for Split (now Harness Feature Management and Experimentation), a feature flag and experimentation platform. Covers feature flag lifecycle, targeting, treatments, segmentation, and experimentation concepts. version: "1.0" created: "2026-05-02" modified: "2026-05-02" terms: - term: Feature Flag definition: >- A toggle mechanism that controls whether a specific feature or code path is active for a given user or traffic segment. Also called a "split" in the Split platform. Feature flags enable progressive delivery and experimentation without code deployments. tags: - Core - Feature Management - term: Split definition: >- The platform's term for a feature flag. Each split has a name, a traffic type, and one or more treatments. Evaluated by SDK or Evaluator API to return a treatment. tags: - Core - Feature Management - term: Treatment definition: >- A named variant that a feature flag can return (e.g., on, off, v1, v2, control). Treatments can carry dynamic configuration (JSON) to adjust behavior without redeployment. tags: - Core - Experimentation - term: Workspace definition: >- A project-level namespace in Split that organizes feature flags, environments, and segments. Corresponds to a business unit, application, or team boundary. tags: - Organization - Administration - term: Environment definition: >- A deployment context (e.g., staging, production) within a workspace where feature flags can have independent definitions, targeting rules, and rollout configurations. tags: - Organization - Deployment - term: Targeting Rule definition: >- A conditional expression that matches a subset of traffic and assigns treatments based on attribute values, segment membership, or percentage-based bucketing. Rules are evaluated in order; the first matching rule determines the treatment. tags: - Targeting - Rollouts - term: Segment definition: >- A reusable group of identity keys (e.g., user IDs) used in targeting rules. Segments can be managed via CSV upload or individual key management operations. tags: - Targeting - Identity - term: Traffic Type definition: >- Defines the entity kind that a feature flag evaluates against (e.g., user, account, device). Each feature flag is associated with a traffic type at creation. tags: - Targeting - Identity - term: Traffic Allocation definition: >- The percentage of incoming traffic that participates in feature flag evaluation (0-100%). Traffic outside this allocation always receives the default treatment. tags: - Rollouts - Experimentation - term: Default Treatment definition: >- The treatment returned when no targeting rules match a given key, or when the feature flag is killed. Acts as the safe fallback for all unmatched traffic. tags: - Core - Feature Management - term: Kill definition: >- An emergency operation that overrides all targeting rules and forces all evaluations to return the default treatment. Used as an immediate feature shut-off. tags: - Operations - Safety - term: Change Request definition: >- A formal request to modify a feature flag definition that requires approval from designated reviewers before being applied. Enables controlled governance workflows. tags: - Administration - Governance - term: Impression definition: >- A record of a feature flag evaluation event for a specific key. Impressions are used for monitoring, debugging, and calculating experimentation metrics. tags: - Observability - Experimentation - term: Evaluator definition: >- The Split Evaluator is a standalone HTTP microservice that wraps an SDK to provide REST-based feature flag evaluation for applications that cannot embed an SDK directly. tags: - Service - Integration - term: Admin API definition: >- The Split REST API for programmatic management of workspaces, environments, flags, segments, users, groups, and API keys. Requires Admin API key authentication. tags: - API - Administration - term: SDK definition: >- A language-specific library that embeds Split's evaluation engine locally, synchronizing flag definitions and evaluating treatments with low latency without making per-evaluation network calls. Available for JavaScript, Node.js, Java, Python, Android, iOS, React, and more. tags: - Integration - Client - term: Bucket definition: >- A percentage allocation within a targeting rule or default rule that assigns a specific treatment to a portion of matching traffic. Buckets in a rule must sum to 100%. tags: - Rollouts - Targeting