How accurate is Mitre Bridge?

← Mitre Bridge live page · all crossings

This crossing sits beside a train-describer berth — a signalling section whose occupation Network Rail broadcasts in real time. Virtually every train that actually crosses is seen the moment it fills that berth, and each pass is checked against what this page was showing at the time — so these numbers measure what a visitor really experiences, not an internal model.

These figures score when trains pass, not when the barriers are down. At the crossings we've audited, the public signalling feed carries no barrier-position bit — a systematic search found none there — so we score train passes instead; the live page derives its closure window from the predicted pass using per-type lead and clear offsets. This is a cctv-monitored manual barrier crossing (CCTV): a signaller monitors the crossing, confirms it is clear, and lowers the barriers ahead of each train — so the gap between barriers-down and the train passing depends on when the signaller acts, which these figures don’t observe.

Yesterday
2026-07-22
Last 7 days
2026-07-16 → 2026-07-22
Trains warned aboutOf the trains observed actually crossing, the share this page was warning about — listed in the timeline or shown live on approach — at the moment they passed. 100.0%99.5%
Warnings that came trueOf the closure warnings the page asserted 30 minutes ahead, the share where a train genuinely arrived — judged against both the signalling feed and official train-movement reports, so a dropped sensor reading isn't miscounted as a false alarm. Warnings with no evidence either way (usually cancellations the page correctly withdrew) are excluded rather than guessed. 82.3%85.8%
Typical timing errorThe median gap between the pass time the page was asserting in its final minutes and when the train actually passed. '9 in 10' is the 90th percentile of the same gaps — only 1 train in 10 was further out than this. ±26s (9 in 10 within 74s)±25s (9 in 10 within 82s)
Timing error 15 min outThe same gap, but measured against what the page was saying 15 minutes before each train arrived — the number that matters when you're deciding whether to set off. ±86s (9 in 10 within 528s)±70s (9 in 10 within 313s)
Trains observedReal passes detected in this window at the signalling berth beside the crossing. 162 1083
Warnings verifiedHow many asserted warnings had enough evidence to be judged for the 'came true' figure. Not every warning can be: a warning with no signalling or movement trace at all is left unjudged rather than guessed. 186 915

How the estimate tightens

Signed timing offset of the page’s estimate on the approach — below the line = warned early, above = late. Median in solid; shaded bands hold the middle half and 8 in 10 — last 7 days. Hover a point for detail.

-240s -120s +120s +240s +360s +480s 0 late ▲ early ▼ 30 min out 20 min out 10 min out at crossing

The offset chart replays each observed train’s minute-by-minute prediction history — every tracked train keeps a rolling queue of what the page was asserting on its approach, sealed when it actually crosses. Offset is predicted minus actual: below zero means the page warned early (annoying but safe), above zero late. The bands narrowing toward the zero line as the train closes in is live signalling progressively correcting the schedule. Hover or tap any metric name above for what exactly it counts. Built nightly at 02:00 UTC from complete days.

built 2026-07-23T22:46:25+00:00 · data