Methodology: how we measure punctuality

Every TrainsRadar figure comes from open real-time data. This page explains what we measure, by which rules, what we cannot measure and when the method changed.

Two different measures

Last known delay. For every train and service day we keep the last delay its source published, at any point of the journey. It is the basis of the monthly report (headline figure and rankings) since August 2026.

Delay on arrival at destination. Since 13 November 2026 we also record whether we saw the train reach its final stop, how late it was, and the cancellations the source publishes. It is the basis of “Punctuality at destination” in Statistics and of the report’s under-5, 6 and 15-minute columns.

The two are never mixed. A “last known delay” figure is not comparable with an operator’s official punctuality, which is measured on arrival and often with other rules.

Definitions

  • Train with data: its delay is known and plausible, between 60 minutes early and 120 minutes late. Anything outside that window is discarded as a source error.
  • On time, by our own rule: last known delay under 3 minutes.
  • On time at destination: arrival at the final stop with less delay than the stated threshold. Early arrivals count as on time.
  • Every percentage is calculated over trains with data, never over all trains.
  • Cancellations: only where the source publishes them. They are not part of the punctuality percentages and are shown separately.

Publication rules

  • An operator enters the monthly ranking with at least 3,000 trains with data in the month and data for 30 % or more of its trains.
  • A line enters the ranking with at least 1,000 trains with data in the month.
  • A country counts as having “real-time delay” with data for 30 % of its trains and at least 3,000 trains with data in the month.
  • For punctuality at destination, a country is not published if we have data for less than 30 % of its scheduled trains, and is not ranked if we miss the arrival of more than 20 % of its trains. Below 50 observed arrivals we give no percentage.
  • Each operator’s official thresholds appear next to ours in the monthly report, with their source and how they differ.

Confidence intervals

Each percentage comes with its 95 % confidence interval (Wilson). To decide whether two countries differ, Statistics looks at the interval of the difference (Newcombe), not at whether the two intervals overlap.

Wilson treats every train as independent. Trains on the same day and line are not (one incident delays many), so the real uncertainty is larger: the interval is a lower bound.

What each country’s source publishes

A delay of “0” does not mean the same everywhere. We measured it in the data itself and reviewed it by hand; where a source does not publish delays below a threshold, that threshold is left without a figure rather than showing a misleading number.

Which countries have real-time delay in a given month is stated in the “Coverage” section of each report.

CountryDelay resolutionWhat a 0 meansVerified official threshold
Australia AUsecondsmeasured: on time –
Bulgaria BGwhole minutesmeasured: on time –
Canada CAwhole minutesunknown (left out of comparisons) –
Switzerland CHsecondsmeasured: on timeSBB: < 3 min
Czechia CZsecondsmeasured: on time –
Germany DEsecondsmeasured: on timeDeutsche Bahn: ≤ 5:59 min
Deutsche Bahn: ≤ 14:59 min
Spain ESsecondsmeasured: on timeRenfe: < 5, < 15, < 30 y < 60 min
Finland FIsecondsmeasured: on timeVäylävirasto: ≤ 5 min (long distance); < 3 min (commuter)
France FRonly incidents ≥ 5 minno incident reportedSNCF: < 5 min
SNCF: < 5 min (< 1h30), < 10 min (1h30–3h), < 15 min (> 3h)
United Kingdom GBwhole minutesmeasured: on timeORR / Network Rail: < 5 min (London & South East, Regional, Scotland); < 10 min (Long Distance)
ORR / Network Rail: < 1 min, < 3 min, < 15 min
Ireland IEwhole minutesmeasured: on timeIarnród Éireann: ≤ 10 min; ≤ 5 min
Italy ITwhole minutesmeasured: on timeRFI: ≤ 5 min (%OS), ≤ 15 min (%OSML)
Japan JPwhole minutesmeasured: on time –
Netherlands NLsecondsmeasured: on timeNS: < 3 min (< 180 s)
Norway NOsecondsmeasured: on timeBane NOR: ≤ 3:59 min; ≤ 5:59 min (langdistanse)
Poland PLwhole minutesmeasured: on timeUTK (Urząd Transportu Kolejowego): ≤ 5 min 59 s
Sweden SEsecondsmeasured: on timeTrafikverket: ≤ 5:59 min
United States USsecondsmeasured: on time –

Sources and licences

The licence matrix lists 75 rail data sources from 32 countries: 34 with a licence we have read that allows commercial use, and 41 still to be confirmed.

The press version of the report only includes countries whose sources allow commercial use.

What we do not measure

  • Passengers: we count trains, not people. A full train weighs the same as an empty one.
  • Causes: we do not attribute delay to a cause. The only exception is an approximation in Finland and Great Britain (time lost behind another train), which is not the official attribution.
  • Trains outside open data: if an operator publishes no real-time data, its trains have no data and are not part of the percentages.
  • Estimated positions: in some countries the map estimates the position of trains without GPS from the timetable.
  • Incomplete days: if data collection fails for part of a day, that day has fewer trains with data. The report’s best and worst day ignore days with less than a tenth of the average volume.

Changes of method

Every change that affects the figures, with its date.

  1. Germany joins the daily record

    The per-train daily record includes German trains from 14 November 2026. The August, September and October reports have no German trains, and therefore no CC BY-SA press file either.

  2. Change of method of 13 November 2026

    From that day we record the delay on arrival at destination and the cancellations each source publishes. Before it, the only figure available is the last known delay along the journey. Comparisons with official figures only use complete months from then on; the first is December 2026.

  3. Daily punctuality by line

    Daily figures by line start. The line rankings of August 2026 only cover 26 to 31 August.

  4. First monthly report

    August 2026 is the first month with a report.

  5. One record per train and day

    We start keeping one record per train and service day, with its last known delay.

How to cite

Cite the specific report with its month and address. Citation formats are on the press page and at the foot of each report.