Experiments Portlet — Screen 3: View Results #37004

Description

The Results screen on the portlet's stack: stat strip, Daily/Bayesian tabs, summary table, promote. The current reports screen is the cheapest to port — its store is already page-independent (state is only {experiment, status, results}, loaded by experimentId alone) — but it is rewritten on Signal Store Events like the rest of the portlet, and the charts/data plumbing are reused, not reinvented.

Heads-up before starting: experiment results still go through CubeJSClient on main (2 CubeJS round-trips + a 1000-sample Monte Carlo per call, @NoCache) — #36763 Part 1 migrates that to CAEM + ClickHouse. Confirm with the analytics team whether to build against the current contract or wait for the new one; the result model (ExperimentResults, GoalResults, VariantResults) is unchanged by the migration, so building now is expected to be safe.

Design: approved prototype, results screen.

Scope

• Route: /experiments/:experimentId/results (placeholder wired by Experiments Portlet — Screen 1: portlet base + site-wide List #36989).
• Header: back to the list · name + status tag · subline {pageTitle} · {pagePath} · {n} Variants · Configuration button (→ Experiments Portlet — Screen 2: Create/Update (/experiments/new + /:id/configuration) #37003's screen) · Stop Experiment when RUNNING.
• Stat strip: Winner (ENDED) / Leading Variant (otherwise) with inline Promote when the leader is not the control · Goal name · Period · Sessions To Date. Ported: "no winner yet" negative state (block icon + legend — the design has no negative state) and the refresh-results button (results are @NoCache; users watch this screen while the test runs).
• Tabs: Daily Results (conversion rate by day, one series per variant, interactive legend) · Bayesian Results (posterior distribution). Reuse the existing Chart.js setup (dot-experiments-reports-chart/chartjs/ chart options) — do not hand-roll the prototype's SVG. Bayesian summary data comes from the backend (bayesianResult in the results payload): conversion rate, credibility interval, probability and risk per variant. The posterior curve is sampled on the client. BayesianResult.distributionPdfs can supply it, but it is gated behind INCLUDE_BETA_DISTRIBUTION_SAMPLES (off by default) and carries BETA_DISTRIBUTION_SAMPLE_SIZE points per variant — a thousand out of the box — on an endpoint that is @NoCache. Sampling a Beta locally costs less than putting those points on the wire on every read.
• Summary table (under both tabs): Variant (dot, name, CONTROL/LEADING chips) · Sessions · Conversions · Conversion Rate · Lift vs Original (percentage points, +11.7 pts, green/red, — on control — the one design addition over the current table) · Probability To Be Best · Conversion Rate Range (95%) · Promote. Ported: minimum-10-sessions gate (below it, empty state instead of meaningless rates) and promoted state (Promote buttons hidden once a variant is promoted; a variant must not be promotable twice).
• Promote confirm: promoting while RUNNING auto-ends the experiment (ExperimentsAPIImpl.java:1392) — the confirm must say so; the design draws a bare button.
• Analytics health gate on this route only: reuse dotAnalyticsHealthCheckResolver from @dotcms/ui (not the edit-page-coupled AnalyticsAppGuard). The list must not be gated.
• Store: Signal Store Events (experiments-results.events.ts), load/refresh/promote as Request → Succeeded → Failed triples.
• Unhide from Screen 1: the View Results primary row action in Experiments Portlet — Screen 1: portlet base + site-wide List #36989 routes here.

Acceptance Criteria

• /experiments/:id/results renders stat strip, tabs and summary per the design for RUNNING and ENDED experiments.
• Daily and Bayesian charts render with the existing Chart.js config; legend toggles series.
• Lift vs Original shows points vs the control row, green/red, — on control.
• Below 10 sessions the summary shows the empty state.
• Refresh re-fetches results without a full navigation.
• No suggested winner → the negative state renders, not a false "Leading Variant".
• Promote asks confirmation; RUNNING copy states the experiment will be ended; after promoting, Promote buttons disappear and the promoted row is marked.
• A misconfigured analytics app blocks only this route, not the list.
• The old UVE reports screen is untouched.
• Jest specs for the results store and summary table.

Priority

Medium — independent of #37003; can run in parallel once #36989's shell exists.

Additional Context

• Depends on: Experiments Portlet — Screen 1: portlet base + site-wide List #36989. Independent of Experiments Portlet — Screen 2: Create/Update (/experiments/new + /:id/configuration) #37003.
• Out of scope: any change to the results backend contract; charts in the list; bayesHeadline/bayesNote narrative copy (present in prototype logic, unrendered — confirmed dropped).
• Port sources: dot-experiments-reports/ (store shape dot-experiments-reports-store.ts:51-55,208, summary, details table, chartjs/).
• E2E (empty state, 10-session gate, promote confirm) lives in the dedicated E2E issue.