The Experiment Timeline and Activity Log

An experiment is only interpretable if you know what it was doing while it ran. The timeline and the activity log are the two records Pertento keeps for that, and they are the first place to look when a chart has a step in it that the change you tested cannot explain.

The timeline

The timeline sits on the experiment page and shows the run as a sequence: when it was created, when it started, and when it ended or was deployed. It answers the question that most often invalidates an analysis — "how long was this actually live" — without you having to reconstruct it from dates in different places.

The activity log

The activity log lives on the Settings tab and records changes made to the experiment’s configuration while it existed: variants added or renamed, traffic weights changed, targeting rules edited.

This matters more than it sounds. A mid-run configuration change makes the data before and after it non-comparable, and a conversion rate that shifts on the day a variant was added is usually explained by the variant being added, not by visitor behaviour.

Using them when results look wrong

When a result surprises you, read the log before you read the statistics again:

1. 1Check whether a variant was added or paused during the run. Traffic shares move when it happens, and the affected period is not comparable to the rest.
2. 2Check whether targeting changed. A widened URL rule pulls in a different audience, so the visitors before and after are not the same population.
3. 3Check whether weights were changed. The sample sizes will not match the split you think you configured.
4. 4Compare against the traffic split check on the Statistics tab. A configuration change is a common explanation for a split that does not match its weights.

Notes

• The safest experiment to interpret is one whose configuration never changed while it ran. Where you can, finish a test and iterate in a new one rather than editing it live.
• The log records configuration changes, not visitor activity. Results live on the Statistics tab.
• Both records survive the experiment ending, which is what makes an old result re-readable months later.