Concepts
Assignment and sticky bucketing
Deterministic bucketing keeps users assigned consistently across SDKs.
Assignment uses deterministic hashing so the same subject, experiment, salt, and variant weights produce the same decision everywhere.
Key properties:
- stable across SDK implementations,
- covered by frozen conformance vectors,
- supports sample rates and variant weights,
- avoids reassignment drift when deployments change,
- sticky client storage keeps browser users stable across reloads.
The web SDK fetches a manifest, computes the decision locally, stores the sticky assignment, and emits an exposure when activation occurs.