Project ID of the project you're trying to access. To find the ID of the project, make a call to /api/projects/.

Name of the experiment.

Description of the experiment hypothesis and expected outcomes.

Unique key for the experiment's feature flag. Letters, numbers, hyphens, and underscores only. Search existing flags with the feature-flag-get-all tool first — reuse an existing flag when possible.

Feature-flag config for the experiment, in the flag's own filters shape. The linked flag is the source of truth for variants, rollout, aggregation, payloads, and experience continuity: send config here instead of the deprecated `parameters` keys. On a running experiment, also send `update_feature_flag_params=true`. Cannot be combined with the key of a pre-existing feature flag on create (the experiment links to it as-is).

ID of a holdout group to exclude from the experiment.

Experiment parameters JSON. Supported keys include `custom_exposure_filter` and `variant_notes` (free-text notes per variant, keyed by variant key). Flag config (variants, rollout, aggregation, payloads, experience continuity) belongs on the `feature_flag` object; send it there. For backward compatibility, config still sent through these deprecated keys is copied onto the linked flag rather than rejected, and reads project the flag's current config back into this field. Excluded variants live on the top-level `excluded_variants` field, not here.

Running-time calculator state: `minimum_detectable_effect`, `recommended_running_time`, `recommended_sample_size`, and `exposure_estimate_config`. Canonical home for these keys, which historically lived in `parameters`.

Variant keys to exclude from metric result calculations. Excluded variants are still served to users but omitted from statistical analysis. The baseline variant and holdout pseudo-variants cannot be excluded. Canonical home for what historically lived in `parameters.excluded_variants`.

IDs of shared saved metrics to attach to this experiment. Each item has 'id' (saved metric ID) and 'metadata' with 'type' (primary or secondary).

Whether the experiment is archived.

Experiment type: web for frontend UI changes, product for backend/API changes. * `web` - web * `product` - product

Exposure configuration including filter test accounts and custom exposure events.

Primary experiment metrics. Each metric must have kind='ExperimentMetric' and a metric_type: 'mean' (set source to an EventsNode with an event name), 'funnel' (set series to an array of EventsNode steps), 'ratio' (set numerator and denominator EventsNode entries), or 'retention' (set start_event and completion_event). Use the read-data-schema tool with query kind 'events' to find available events in the project.

Secondary metrics for additional measurements. Same format as primary metrics.

Suppresses the validation that rejects metrics referencing events not yet ingested by this project. REQUIRES explicit user confirmation before being set to true — never flip this silently to retry a failed call. The default validation catches typo'd event names and missing instrumentation. Set this to true only when the user has confirmed the event is intentional (e.g. they are about to instrument it).

Experiment conclusion: won, lost, inconclusive, stopped_early, or invalid. * `won` - won * `lost` - lost * `inconclusive` - inconclusive * `stopped_early` - stopped_early * `invalid` - invalid

Comment about the experiment conclusion.

When true, sync the flag config sent in this request (via the `feature_flag` object) to the linked feature flag. Draft experiments always sync regardless. On a running experiment, `feature_flag` config without this flag is rejected.

Experiment lifecycle state: 'draft' (not yet launched), 'running' (launched with active feature flag), 'paused' (running with feature flag deactivated — virtual state derived from feature_flag.active, not stored), 'exposure_frozen' (running with enrollment frozen to the already-exposed cohort while metrics keep flowing — virtual state derived from the flag's release groups, not stored), 'stopped' (ended).

Whether the experiment uses any legacy-engine metrics (ExperimentTrendsQuery or ExperimentFunnelsQuery). Used to flag legacy experiments and gate actions that don't support them, such as duplicate and copy-to-project.

The effective access level the user has for this object