FabricFabricExperiments
Concepts

Statistics

Exposure-first frequentist, Bayesian, CUPED, timing, depth, and segmented experiment results.

Fabric Experiments starts with pragmatic, auditable statistics.

Aggregate rows

The aggregate endpoint returns per-variant rows:

  • variantKey
  • exposureCount
  • conversionCount
  • conversionSum
  • failureCount

The response envelope adds experimentId, metric, computedAt, and optional analytics. Raw rows are counts and sums; they do not contain Bayesian summaries. The statistics package and Studio derive Bayesian results separately from those sufficient statistics.

The endpoint accepts metric plus optional ISO fromIso/untilIso bounds, segmentKey, and prePeriodDays (1–365):

GET /api/v1/orgs/{orgId}/experiments/{experimentId}/aggregate?metric=signup&fromIso=2026-07-01T00%3A00%3A00.000Z&untilIso=2026-07-08T00%3A00%3A00.000Z&segmentKey=country&prePeriodDays=14

Result methods

Fabric reports frequentist rates, lift, intervals, p-values, and sample-ratio mismatch checks. Bayesian results use a documented Beta-Binomial model with a Beta(1,1) prior. CUPED variance reduction is available when the configured pre-period covariate has usable variance.

Studio also surfaces median and P90 time-to-convert, 1+/2+/3+ conversion depth, and results for explicitly allow-listed low-cardinality segments.

These are fixed-horizon analyses. Always-valid sequential inference is not currently provided. See Read experiment results for interpretation and data requirements.

On this page