Remote config, flags & experiments
How parameters, feature flags, experiment assignment and exposures work.
Every SDK calls POST /v1/sdk/config with the user’s pseudonymous ID and consent, and caches the result (default TTL 300 seconds) for offline use.
{
"parameters": { "onboarding_steps": 2, "difficulty_multiplier": 1.0 },
"flags": { "new_dashboard": true },
"experiments": [{ "experiment_id": "e1f…", "parameter": "onboarding_steps", "variant": "short" }],
"experiences": { "daily_mission": { "variant_id": "v9…", "version": 3, "content": { … } } },
"actions": [{ "campaign_id": "c2…", "type": "offer", "payload": { … } }],
"ttl_seconds": 300
}Parameters
Define parameters in the dashboard with a type (number, integer, boolean, string, enum), bounds or allowed values, and whether Autopilot may tune them. Values outside bounds are rejected by the API — for humans and AI alike.
Assignment
Assignment is deterministic and sticky: sha256(experiment_id:user_id) mod 10000. Users who have not granted personalization consent always receive default values and are never enrolled in experiments.
Exposures
When your code reads a parameter that is under test (getParam), the SDK sends one $exposure event per experiment per session with { experiment_id, variant }. Results are computed on exposed users only, within the experiment’s attribution window (default 14 days).
Measurement
Results include conversion and revenue per user per variant, two-proportion and Welch tests, 95% confidence intervals, a sample-ratio mismatch check and a count of users exposed to more than one variant (excluded).