Checked against the product’s code on 7 November 2026.
Under Optimization > Personalization, the tabs What it has learned, What it is trying and What needs a person show recorded evidence and policy. They do not establish that an automatic cycle of testing and applying changes is running on your site.
Read the recorded evidence
Section titled “Read the recorded evidence”- Open What it has learned.
- Inspect a reading for the intended slot and piece of content.
- Read its evidence and context.
- Check how much the configuration is willing to trust that reading.
When there is too little evidence for one set of circumstances, a reading from a broader group can be used.
The default trust setting is zero. A screen containing a lift reading therefore does not by itself mean that reading is changing delivered scores. Review the active settings and connection with Amadalis before depending on that effect.
Review rules for trying alternatives
Section titled “Review rules for trying alternatives”Exploration defaults off.
- Open What it is trying.
- Review the rules for trying alternatives.
- Review the share of traffic they use.
- Review what they may draw from, for how long and within which bounds.
An envelope is a saved set of limits for possible changes. Record this envelope is a deliberate action with its own consequence and reason. Recording it is not evidence that a cycle of automatic changes has started or that any page was published.
Keep a person’s decisions visible
Section titled “Keep a person’s decisions visible”What needs a person counts what is waiting on a human: visitor erasures not yet carried through, campaigns a person is holding, slots set up but never served, and saves the platform refused.
- Open What needs a person.
- Open the count for the list you want to inspect.
A held targeted item is resumed on Targeted content under Personalization. To decide a proposal to change the engine’s settings:
- Open What it is trying.
- Choose Apply it or Reject it for that proposal.
- Read its approval card.
- Approve only the intended decision.
- Check the exact result and current state afterwards.
Continue with slot and measurement settings for delivered ranking rules, or affinity settings for the decaying scores of visitor actions.