Experiment Bulk Results API

GET /api/v1/experiments/{id}/bulk-results returns every successful snapshot for one experiment generated inside a date window, flattened into ExperimentBulkResult items. Use it to mirror historical results into a warehouse or notebook; use GET /experiments/{id}/results when you only need the latest analysis.

info

Off by default and excluded from the REST API reference. Self-hosted deployments opt in with EXPERIMENT_BULK_RESULTS_ENABLED=true. While disabled, every request answers 404 {"message": "Unknown API endpoint"} — indistinguishable from an endpoint that does not exist. This choice is because this endpoint can possibly consume a lot of resources from the back-end deployment if the requested payload is enormous, so it should be used with caution.

Request

Authenticate with a Personal Access Token or secret key that can read the experiment's Project.

curl -X GET 'https://api.growthbook.io/api/v1/experiments/exp_abc123/bulk-results?dateStart=2026-01-01T00:00:00Z&dateEnd=2026-02-01T00:00:00Z' \
  -H 'Authorization: Bearer YOUR_API_KEY'

Param	Required	Description
dateStart	yes	ISO 8601 date-time. Snapshots generated on or after this instant.
dateEnd	yes	ISO 8601 date-time. Snapshots generated on or before this instant.
phase	no	Zero-based phase index. Omit for all phases.
snapshotType	no	standard, exploratory, or report. Omit for all types.
limit	no	Snapshots per page, 1–100. Defaults to 10.
offset	no	Snapshots to skip. Defaults to 0.

The query is strict: an unrecognized param is a 400. Other failures are 400 for a malformed dateStart/dateEnd/phase or an out-of-range limit, 404 for an unknown or unreadable experiment id, and 429 past the per-key cap (60 requests/minute, set by EXPERIMENT_BULK_RESULTS_RATE_LIMIT_MAX).

Response

{
  "results": [
    {
      "id": "snp_abc:overall",
      "snapshotId": "snp_abc",
      "experimentId": "exp_abc123",
      "phase": "0",
      "type": "standard",
      "triggeredBy": "schedule",
      "dateCreated": "2026-01-14T04:11:02.000Z",
      "dateStart": "2026-01-01T00:00:00.000Z",
      "dateEnd": "2026-01-14T00:00:00.000Z",
      "dimension": { "type": "none", "precomputed": false },
      "settings": {
        "datasourceId": "ds_abc",
        "assignmentQueryId": "user_id",
        "experimentId": "homepage-cta",
        "segmentId": "",
        "queryFilter": "",
        "inProgressConversions": "include",
        "attributionModel": "firstExposure",
        "statsEngine": "bayesian",
        "regressionAdjustmentEnabled": true,
        "goals": [
          {
            "metricId": "met_signup",
            "effectiveSettings": {
              "windowType": "conversion",
              "windowValue": 72,
              "windowUnit": "hours",
              "delayValue": 0,
              "delayUnit": "hours",
              "properPrior": false,
              "properPriorMean": 0,
              "properPriorStdDev": 0.1,
              "regressionAdjustmentEnabled": true,
              "regressionAdjustmentDays": 14
            }
          }
        ],
        "secondaryMetrics": [],
        "guardrails": []
      },
      "results": [
        {
          "dimensionValue": "",
          "totalUsers": 41230,
          "checks": { "srm": 0.62 },
          "metrics": [
            {
              "metricId": "met_signup",
              "metricName": "Signups",
              "variations": [
                {
                  "variationIndex": 0,
                  "variationKey": "0",
                  "variationId": "var_control",
                  "variationName": "Control",
                  "users": 20604,
                  "analyses": [
                    {
                      "engine": "bayesian",
                      "differenceType": "relative",
                      "numerator": 1834,
                      "denominator": 20604,
                      "mean": 0.089,
                      "stddev": 0.284,
                      "effect": null,
                      "ciLow": null,
                      "ciHigh": null,
                      "chanceToBeatControl": null
                    }
                  ]
                }
              ]
            }
          ]
        }
      ]
    }
  ],
  "limit": 10,
  "offset": 0,
  "count": 1,
  "total": 37,
  "hasMore": true,
  "nextOffset": 10
}

Item fields

Field	Description
id	Unique per item: {snapshotId}:overall, or {snapshotId}:dimension:{urlEncodedDimensionId}.
snapshotId	Shared by all items expanded from the same snapshot.
experimentId	Internal experiment id. Note settings.experimentId is the experiment tracking key instead.
phase	Zero-based phase index, as a string.
type	standard, exploratory, or report. reportId is present when report.
triggeredBy	What ran the snapshot, when recorded. Either manual (manually triggered by a user in the UI or API), schedule (triggered by the scheduled cron job), manual-dashboard (triggered by a user in an Experiment Dashboard), or update-dashboards (triggered by the update job that refreshes Experiment Dashboards).
dateCreated	ISO date-time the snapshot was generated. Snapshots are returned newest first.
dateStart / dateEnd	Analysis window frozen at snapshot time.
dimension	{ type, id?, precomputed }. type is none, experiment, user, or a pre-exposure kind such as date.
settings	Snapshot-authoritative analysis settings, including per-role metric lists (goals, secondaryMetrics, guardrails, activationMetric).
results	One entry per dimension slice: dimensionValue, totalUsers, checks.srm, and metrics.

Each metric role entry is { metricId, effectiveSettings? }, where effectiveSettings is the snapshot's computed metric settings (conversion window, delay, prior, regression adjustment, targetMDE). It is omitted for snapshots generated before those were captured.

Variation analyses

Each variation carries variationIndex and variationKey — the authoritative join keys, taken from the snapshot's stored variation array — plus its users count and an analyses array holding one entry per stored difference type: engine, differenceType (relative, absolute, scaled), numerator, denominator, mean, stddev, effect, ciLow, ciHigh, pValue, and chanceToBeatControl. Every statistic is null when it is missing or non-finite, and effect is expressed according to differenceType.

Semantics

Snapshot-authoritative. settings, the metric lists, effectiveSettings, the analysis window, and each variation's variationKey/variationIndex come from the stored snapshot and the analysis that produced the numbers. variationId, variationName, and metricName are best-effort current display metadata resolved by id, so they may reflect later renames or be absent entirely. Legacy snapshots fall back to current experiment values only where the stored field is absent.

One item per dimension. A snapshot expands into one item per dimension, all sharing snapshotId. Only the snapshot's default analysis and its stored difference-type variants are returned; missing variants are not computed on the fly, and ad-hoc analyses with other settings (a different baseline or stats engine) are excluded.

Pagination is over snapshots. total counts matching snapshots and count counts the snapshots on this page, so a page of count: 1 can still return many results items. hasMore and nextOffset advance over snapshots too. Offsets are only stable once the window has closed: a snapshot generated inside the window while you page through it sorts ahead of older ones and shifts later pages.