Your completed tests become learnings.

In detail.

What a test holds

Name, area, themes, dates, primary metric, effect and range, result, decision (rolled out, original kept, abandoned), and what the team learned, in one sentence.

Result

Computed by the engine, as for a running test, and always read with its metric: winner, loser, or no detectable effect. A test with a sample ratio mismatch appears as “skewed measurement”, excluded from learnings and priors.

Write-up

Drafted by Probant from the report, with the same figures: the measured and corrected differences each named, and no direction claimed when the range includes zero. The team reviews it, edits it, and publishes it to the library.

Imported tests

Tests from AB Tasty, VWO, Kameleoon, Optimizely, Convert, or GrowthBook join the list with their tool’s name, recalculated from their raw data; they count toward the quality score and the priors.

In the test plan

A new test’s target effect is placed among the effects of completed tests: realistic, ambitious, or never measured in the store.

Sharing and export

“Invite as viewer” creates an invitation for a member with the Viewer role; the CSV export includes the list, with effects in percentage points, and every export is recorded in the audit log.

Pooled effect

Average of the theme’s effects weighted by the inverse of their variance, with its range.

Confidence

Low below 3 tests; high if the pooled effect excludes zero; medium otherwise.

Prior

Mean and spread of true effects, estimated across all completed tests (48 in the demo), accounting for each one’s uncertainty.

Search

Across names, areas, themes, and what the team learned, ignoring accents, even phrased as a question (“What have we tested on pricing?”); a link to the library can carry its question.