We name seven possible causes, and always show what we do not yet know.

The causes

• Operator not available, vehicle, schedule too short, road conditions, dispatch and bunching, and external events.
• Not yet known. This one is always shown.

Each cause carries its basis: stated by an operator notice, inferred from patterns, or what is left over. It also carries a confidence, the tests it ran, whether each passed, and the value we measured, and links to the evidence.

How a cause is chosen

• Missing trips: the notice’s own reason wins. Then a breakdown alert around the same time. Then weather and event days beyond the spread of ordinary days of the same type. Then the operator pattern, such as block starts or one garage.
• Late visits: weather days first, then a fixed slow place, then a schedule that is too short, then bunching on frequent routes. Claims that cannot all be true are scaled down together. The rest is not yet known.
• Ranges are sampling ranges, added across hours, so they are wide on purpose.
• The operator’s published reason mix is shown as evidence only. It never explains a single route.

The words are written by people

The sentences on the why pages are written by people, in English and French. A model may only choose which written sentence leads, which caveats stay, and one closing check. It never writes words or figures. A sentence with a wrong number is rejected before anyone sees it.

Where the numbers come from

• Our measurements
• Operator service notices
• Weather and event days

Method version and change log

A published month never changes. If a method is revised, past months keep the method they were worked out with.

• Version 1 · October 25, 2026. First published. Seven causes, one of them always "not yet known".