How Pertento Decides What Needs Your Attention

Rather than leaving you to review every running experiment, Pertento works out which ones need a person and puts them in order. The list appears on your dashboard and on the Health page, grouped into three bands.

The three bands, in order

Broken — something is wrong with the experiment itself. *Do not act on these results.* They come first because everything below them is a decision, and a decision made on faulty data is worse than a decision postponed.

Needs a decision — these have reached a conclusion and are waiting on your call.

Queue — work that is ready to pick up. Nothing is wrong and nothing is urgent.

What lands in Broken

• Did not split as configured — the observed traffic split is further from the configured weights than chance explains. Every result on the experiment is suspect until the cause is found.
• Stopped receiving traffic — nothing recorded for the last day while the experiment is still running. Usually a targeting rule that stopped matching, or the script missing after a site deploy.

What lands in Needs a decision

• Has a winner — a variant is ahead by more than chance explains. Roll it out, or keep running for a tighter interval.
• Is losing — a variant is behind the control by more than chance explains. Every further day costs conversions, which is why this is not left to be noticed later.
• Is flat — sequential testing has ruled out the effect the test was sized to find. Stopping frees the traffic for the next idea, and that is a real outcome rather than a failure.
• Past its end date — scheduled to finish but still running and still splitting traffic. End it or extend the schedule.

What lands in Queue

• Won but was never built — a winning variant that is not live and has not been marked implemented. A winner that never ships is worth nothing.
• Still served by Pertento — rolled out and working, but not yet permanently coded. It functions; the change belongs in your codebase.
• Next up — when nothing above needs you, Pertento suggests your highest-scoring draft, with its PXL score. On a quiet day the platform answers "what should I work on" instead of leaving you to re-read the backlog.

How to work the list

Top to bottom. Fix what is broken before deciding anything, decide what is waiting before starting anything new, then pick up the queue. Each row carries the action it wants — Investigate, Check setup, Review, See result, Open queue — and links straight to the page where you do it.

An empty list is a real state and a good one. It means nothing is broken, nothing is waiting, and nothing is stuck.

Notes

• The same signals are delivered as notifications, so you do not have to be looking at the dashboard to find out.
• Health shows these across every account you can reach; the dashboard shows them for the selected website.
• New signals are added over time. An item you do not recognise still names the experiment and links to it.