A thousand messages about one delay, and the twelve that mattered.

A single signal failure generated over nine hundred messages in ninety minutes. The operations team read the first forty, concluded it was about the delay, and moved on. Everything genuinely new in that corpus stayed unread.

The network also had an accessibility problem it could measure only by absence. Step-free access reports arrived rarely, from the people least likely to report twice, so every volume-ranked dashboard treated the issue as settled.

Incident clustering was configured first. Each disruption event collapses into a single tracked entity, with distinct reports surfaced separately, so operationally significant submissions are not buried under high-volume duplicate complaints.

Accessibility was broken out as a first-class category with its own escalation floor and its own baseline, reported monthly whatever the volume.

Route, station and shift baselines went in, so a deteriorating branch line stays visible beside a busy central terminus.

Inside the first month the console showed that nine of eleven delay reports described being left uninformed rather than being delayed. The fix moved from engineering to customer information, and it cost a rota change.

Accessibility reports, measured against their own baseline, showed a sustained rise at three stations no volume ranking had ever surfaced. All three sat with the same maintenance contractor.