Methodology
How Delay board turns raw train reports into its rankings. Every operator is measured with the same rules, as far as its source allows. Where a source cannot meet a rule, the board says so instead of hiding it.
A train’s delay
A train’s delay is its latest known delay. Delay board keeps one record per train, and each newer report replaces the one before, so only the latest counts. A train that runs early counts as 0 minutes late, never as negative.
The delay is taken at the last stop the train has passed for which its source gives a time, and stored as actual. Before the train reaches such a stop, the forecast for its next stop stands in, stored as current. Deutsche Bahn’s Timetables API, the source for long-distance trains in Germany, does not say whether a time is actual or forecast, so its delays are all stored as current. Each operator page marks which kind each of its listed trains is.
Lines today
The front page ranks today’s lines in Germany, the Netherlands, Switzerland, Belgium, Norway and Sweden: each line’s trains due to set off since midnight (the six countries share one clock). Choose one country to rank its lines on their own. A line is an operator’s line in one regional network. Line names repeat, so each line is shown with its operator and its two ends.
Lines are the names riders see: regional lines (RE 42, RS18, S 3), DB’s long-distance line numbers (ICE 10), NS’s train series (IC 2800), in Belgium the lines Infrabel, its railway network, gives them (IC 06, S1), in Norway the operators’ line numbers (R10, F6), and in Sweden the numbers of Stockholm’s Pendeltåg (41, 43X), else the train’s product between its two ends (Pågatågen Malmö C – Ystad). Trains without a line, such as Eurocity trains in Switzerland, count on the Companies tab only.
A train counts as delayed when its latest known delay is 5 minutes or more. A line’s “% delayed” is its delayed trains divided by the trains that ran. Cancelled trains are counted separately and are not part of it. Trains cancelled for planned works count as cancelled, because for a rider they do not run.
“Delay today (total)” on every row adds up the minutes all of the line’s trains were late, so you can see how much delay the line has stacked up so far that day. “Per train” beside it divides that total by the trains that ran, so you see how late they were on average; cancelled trains are not counted.
On a country’s own board every line is ranked from 2 trains, by its delayed share, so the board ranks its lines as soon as they run; a line with few trains can top it. On the board of all countries, and in each day’s final ranking kept for good (the day pages, the Delay MVP, the year), a line is ranked once 10 of its trains have run, so a few late trains cannot decide it. In Belgium’s own final ranking, as its feed shows only about one running train in five, 2 trains count as 10. The year’s “Highest % delayed” list takes only lines with that many trains for every day counted. Lines are ranked by their delayed share as the board prints it, in whole percent, highest first; ties go to the higher average delay per train, then to more trains run.
On time
A train is on time when it is less than 5 minutes late. The threshold is a setting and applies to every operator alike.
On the company tab, the on-time rate is on-time trains divided by all trains, cancelled trains included. Many operators publish their own punctuality figures with different thresholds, so the board’s rates will not match theirs.
Cancellations
On the company tab, a cancelled train counts as not on time, so cancellations lower the on-time rate. It is left out of the average delay, because a train that never ran has no delay to average. On the line board, cancelled trains are counted separately (see Lines today).
Windows
The company tab uses rolling windows rather than calendar days. A train belongs to a window when it was last seen inside it: a running train at each report, a finished one at the last stop with a time, a cancelled one at its planned start.
- Live
- The last 60 minutes.
- 24 h
- The last 24 hours. The company tab opens here.
- 7 days
- The last 7 days.
“Most delayed now” on each row looks only at trains that reported in the last 60 minutes, whatever window is chosen.
Ranking
On the company tab, operators are ranked most delayed first: by average delay, or by on-time rate with the lowest first. Ties go to the other measure, then to the larger sample.
An operator is ranked only with enough trains: at least 10 for Live and 30 for 24 h and 7 days. Below that, it is listed under “Not ranked” with “Too few trains to rank”, and its figures are printed in ink, not in delay red.
Snapshots
Figures are recomputed every 5 minutes. The change column compares each line’s or operator’s rank with the previous snapshot of the same view. Each train’s latest state is kept for 14 days. Each day’s final line ranking is kept for good 3 hours after the day ends at midnight Berlin time, once its late trains have arrived.
When a source stops answering, the rest of the board carries on. Its lines and operators are marked “No data from the source since” the last successful fetch, and their figures cover only the trains reported before then. A country whose collection lapsed for more than 30 minutes at a time between the day’s midnight and its keeping (its source down, or the board stopped), or began only after that midnight, is left out of that day’s final ranking; the other countries keep theirs.
Delay bands
Operator pages sort trains into five bands, drawn in ink density so they read without colour:
- On time: less than 5 minutes late
- 5 to 15 minutes late
- 15 to 30 minutes late
- 30 or more minutes late
- Cancelled
What Delay board is not
Delay board is not official punctuality data. Sources measure delay differently: most give a time at every stop, Switzerland’s gives a delay only where it changes, for Germany’s long-distance trains Delay board reads only a set of hub stations, and Sweden’s holds only Swedish stops, so trains to Copenhagen, Oslo or Narvik are measured as far as the border. Coverage differs too, so compare operators with that in mind.
Each source and what it covers is listed under Sources and licenses.