Experiment overview

A per-studio page rolling up every A/B test report in the studio into one place — running tests, concluded ones, and how they're doing — so you can read the whole portfolio without opening each report individually. Individual A/B test reports keep working in every deployment; this page is the cross-report view across them.

Reach it from the studio's folder bar on the dashboard, or from the "This studio" menu on the crumb — both list it alongside the studio's other feature surfaces, separate from the report folders themselves.

How a report joins the page

Nothing new to build. A report joins the overview the moment its report.yaml declares an ab_test: block — the same block the report's own A/B comparison component already reads for its header (test name, description, split, start date, control/test labels). A report with no ab_test: block simply never appears here.

A report that carries several experiments on one page — a programme review, a portfolio page — declares a list of such blocks instead. Each entry becomes its own row and its own timeline lane here, matched to its own A/B component in the built output by test_name (falling back to declared order against build order when no names are given):

ab_test:
  - test_name: checkout_v2
    start_date: "2026-07-14"
  - test_name: starter_pack_price
    start_date: "2026-07-09"
    end_date: "2026-08-06"

Four additional keys are recognised by this page only — the report's own build ignores them entirely:

• primary_metric — which KPI row the overview should lead with, matched against the row's key or its metric label. Without it, the first KPI row in the report's own payload order is used.
• end_date — when the test is (or was) meant to stop.
• planned_end — an alias for end_date, for authors who'd rather name it as a plan than a commitment. Checked only when end_date is absent.
• closed — the decision was made and the chapter is closed: the entry reads as concluded even while the report keeps rebuilding on a schedule. This is the honest state for a finished test sharing a page (and therefore a schedule) with one still running.

Status rules

Each experiment's status is derived from start_date and end_date / planned_end:

• scheduled — start_date is in the future.
• running — started, and no end date has passed (or none was declared at all — an experiment with no end runs open-ended until its report author adds one).
• concluded — an end date exists and has passed.
• past end — an end date exists and has passed, but the report is still on a live build schedule. That combination means the test outran its own declared horizon and nobody told the pipeline to stop rebuilding it — shown as a nudge, not an error. A report that's disabled, or has no schedule, past its end date is simply concluded — as is any entry that declares closed: true.
• unknown — no parseable start_date; nothing to place on the timeline.

Dates are read as plain calendar days in UTC, regardless of where the studio's members sit.

Where the numbers come from

This page computes no statistics of its own. Every number — the delta, its confidence interval, which variance-reduction mode is showing — is read straight out of the report's own latest built output, at that report's own declared default mode (default_mode in the ab_test component). If a report declares a raw view and a CUPED-adjusted view and opens on the CUPED-adjusted one by default, that's the number shown here too: this page always mirrors the same view you'd land on by opening the report directly.

A lift renders green or red only when its 95% confidence interval excludes zero — and its direction is judged against the metric's own declared "higher is better" — otherwise it renders as not-yet-significant rather than implying a winner the data doesn't support. A "win" counts only a concluded experiment with a significant, favourable result.

A report that has never built shows "Not built yet"; a report that built but whose output has no A/B payload shows as much, rather than disappearing from the list or breaking the page.