Context
Frontend design pass on the Experiments v2 surfaces in the prototype:
the Create Experiment wizard, the Experiments list page, and the
Experiment results page.
Goal
Land a wizard shape, list, and results page where every surfaced control
corresponds to behaviour the platform supports today. Controls that
would be placeholders for behaviour not yet implemented are deferred and
revisited when the backing capability exists. This keeps V1 honest:
no fake choices, no settings that don't change anything.
In scope
-
Create wizard. Step structure, validation, copy, primitives.
-
Audience model. How an experiment targets a subset of identities
using inline attribute conditions, frozen at launch.
-
Eval hierarchy. How the experiment composes with existing flag
behaviour:
- Identity override on the flag
- Segment override on the flag
- Experiment allocation
- Default flag value
First match wins. Identities caught by 1 or 2 do not enter the
experiment; banners on the wizard surface this up front.
-
Metric selection. Roles, defaults, primary-metric semantics, and
inline metric creation.
-
Experiments list. States, columns, empty/error/loading states.
-
Results page. Metric comparison, trend chart, recommendation
callout, summary cards.
Out of scope
- Backend or API changes
- Statistical methodology selection (the platform pins V1 defaults)
- Scheduled launches and auto-stop logic
- Cross-experiment coordination
Deliverables
Done when
Both deliverable PRs are merged.
Context
Frontend design pass on the Experiments v2 surfaces in the prototype:
the Create Experiment wizard, the Experiments list page, and the
Experiment results page.
Goal
Land a wizard shape, list, and results page where every surfaced control
corresponds to behaviour the platform supports today. Controls that
would be placeholders for behaviour not yet implemented are deferred and
revisited when the backing capability exists. This keeps V1 honest:
no fake choices, no settings that don't change anything.
In scope
Create wizard. Step structure, validation, copy, primitives.
Audience model. How an experiment targets a subset of identities
using inline attribute conditions, frozen at launch.
Eval hierarchy. How the experiment composes with existing flag
behaviour:
First match wins. Identities caught by 1 or 2 do not enter the
experiment; banners on the wizard surface this up front.
Metric selection. Roles, defaults, primary-metric semantics, and
inline metric creation.
Experiments list. States, columns, empty/error/loading states.
Results page. Metric comparison, trend chart, recommendation
callout, summary cards.
Out of scope
Deliverables
Done when
Both deliverable PRs are merged.