Section 6 · OES Spec
Analysis configuration
How the scorecard was computed — reproducibility starts here.
A result is not 'conversion +3%'. It is 'conversion +3% under this model, window, population, and variance estimator'. The analysis block makes those choices explicit and machine-readable.
Fields
| Field | Type | Purpose |
|---|---|---|
| analysis.method | enum | frequentist, bayesian, sequential, cuped, diff_in_diff, etc. |
| analysis.model | string | t-test, regression, bootstrap, Bayesian normal, binomial, etc. |
| analysis.varianceEstimator | enum | naive, delta_method, cluster_robust, sandwich, bootstrap. |
| analysis.confidenceLevel | number (0–1) | Usually 0.95. |
| analysis.alpha | number (0–1) | Significance threshold. |
| analysis.prior | object | Bayesian prior, if relevant. |
| analysis.adjustmentMethod | string | CUPED, regression adjustment, stratification, etc. |
| analysis.multipleComparisonCorrection | string | Correction method. |
| analysis.segmentation | string[] | Segments included in analysis. |
| analysis.dimensionBreakdowns | string[] | Browser, country, new/returning, plan type, etc. |
| analysis.missingDataHandling | string | How missing outcomes were treated. |
| analysis.outlierHandling | string | Winsorization, trimming, capping. |
| analysis.queryReferences | object[] | SQL query IDs, warehouse job IDs, notebooks. |
| analysis.generatedAt | string (date-time) | Time the result was generated. |