What experiments are
Experiments are segmented visibility work.
That means an experiment can hold a smaller set of questions, scheduled agent prompts, chatlogs, and dashboard readouts. It is not a separate product. It is a focused view inside the same workspace.
Experiments are segmented visibility work
When an experiment is selected, CLEA reads that experiment as the active scope.
The question panel edits experiment questions instead of the main mention prompts. The calendar can show and edit experiment agent prompts. Analytics, Sources, Competitors, and History use experiment chatlogs and experiment questions when they exist.
Experiment scope
When selected, the experiment narrows questions, chatlogs, analytics, sources, competitors, calendar work, and goal notes.
Why experiments are kept separate
Separate scope can make a test easier to read.
If you are testing a new page cluster, you may not want all website prompts mixed into the result. If you are testing one competitor angle, you may only want questions tied to that angle.
| Area | Experiment behavior |
|---|---|
| Questions | Uses the experiment question set |
| Analytics | Reads experiment prompt and chatlog data |
| Sources | Builds source insight from experiment chatlogs |
| Calendar | Can manage experiment scheduled agent prompts |
| Studio | Edits the experiment goal in `EXPERIMENT.md` |
Current experiment limits
Experiments should not be treated as perfect A/B testing.
CLEA can help isolate questions and scheduled work, but AI answers can move for many reasons. Use experiments to learn from a focused change. Do not use them to claim certainty from one small result.