Munim Technologies

Research/Society

One in five catalogued transit timetables has run out of dates

Of 3,502 public-transport timetable feeds we could read in the world's largest open catalog, 698 (19.9%) had no service left. In Japan, most gaps in published bus timetables open on April 1, and we have recorded a forecast for 2027.

Oct 23, 2026 · 7 min read

Transit apps learn when buses and trains run from GTFS feeds: zip files of timetables that each agency publishes and that catalogs collect. When a feed's dates run out, apps show no service or the wrong service. We read every GTFS feed in the Mobility Database, the largest open catalog. Of 3,502 feeds we could read, 698 (19.9%) had no service left on 22 October 2026, and 392 of those had run out more than a year earlier. US feeds were expired far more often than Japanese ones (28.8% against 7.6%). In Japan, a separate census found that most gaps in published bus timetables open on a single day, April 1, the start of the fiscal year.

19.9%

of 3,502 catalogued feeds read had no service left

96.2%

of expired copies checked are the same file the producer serves

59.6%

of 203 Japanese timetable lapses began on April 1

~47 of 292

Japanese feeds forecast to lapse on 2027-04-01 (range 42-59)

The question

How many of the world's catalogued transit timetables have run out of dates, and is the problem at the source or in the catalog? Validators flag expiry one feed at a time, and one US count exists, but we found no worldwide, country-by-country count. For Japan, which publishes hundreds of small-town bus feeds through a national repository, we asked how often a feed has no valid file at all and when those gaps open.

What we did

For every GTFS entry in the catalog with status active, inactive or future, 3,615 in all, we opened the catalog's copy of the latest file and read only its calendar files. A feed is expired when the last date on which any service runs is before the as-of date. Six predictions were recorded before the run. A second tool then compared each expired copy with the file at the producer's own URL, again with predictions fixed in advance. For Japan, we used the version history that the GTFS Data Repository returns for 602 feeds and counted the days on which a feed had no valid published file, from 2021 to 16 October 2026.

What we found

All six predictions for the world census held. Nearly all feeds the catalog marks inactive have indeed run out (98.0%), and the median inactive feed had ended 510 days earlier. Expiry varies widely by country. The United States, with 1,230 feeds read, sits near the top; Japan, with 629, is among the lowest. The difference is far beyond chance (one-sided Fisher test, p = 4.6e-29).

Share of catalogued GTFS feeds with no service left, by country

Countries with at least 40 feeds read. As of 2026-10-22; statuses active, inactive and future. Source: gtfs-world-expiry/world/2026-10-22-all/summary.md

Is the catalog to blame? Mostly not. Of the 479 expired entries whose producer file we could read, 461 (96.2%) are the very same file: the timetable has really ended at the source. But 17 (3.5%) are stale copies. The producer publishes a newer timetable that still runs, and the catalog has not stored it. They include MARTA in Atlanta, Metro St. Louis, Milwaukee County Transit, Tampere in Finland and Metro de Málaga. A week later none of the 17 had been renewed. We reported them to MobilityData, which maintains the catalog.

Catalog copy against the producer's own file

Expired group (698)Control group (300)
Same file at the producer461239
Producer file differs and still runs17 (stale copies)8
Producer file differs and has ended too10
Producer URL could not be read196 (HTTP 404: 105)32
Producer URL needs a key (not read)2321
Control group: a seeded random sample of entries whose copy still runs. The 8 differing control files were all stored by the catalog within two days and are not stale. Source: gtfs-world-expiry/source/2026-10-23/summary.md.

Japan's April 1 cliff

In Japan's national repository, 28,029 of 492,026 feed-days (5.7%) had no valid published file, across 203 lapses in 156 feeds. The lapsed share rose from 5.5% in 2024 to 6.7% in 2025 and 7.0% in 2026 so far. Of the 203 lapses, 121 (59.6%) began on April 1, when many municipal contracts and timetables turn over with the fiscal year. Among feeds whose file ended on March 31, the share with nothing valid on April 1 reached 20.2% in 2026.

Japanese feeds ending March 31 that had no valid file on April 1

0510152020222023202420252026YearShare lapsed (%)
Feeds ending March 31: 40 (2022), 99 (2023), 150 (2024), 230 (2025), 272 (2026). Shares from lapsed_share in the file, shown as percentages. Source: gtfs-jp-lapses/data/2026-10-16/fy_cliff.csv

These gaps are not brief. Counting lapses still open, the median time until a valid file exists again is 80 days, and 13.4% were still lapsed after a year. Two thirds last longer than the one month for which the Transit app says it extends an expired schedule. Feeds that lapsed on the previous April 1 lapsed again in 2026 in 10 of 25 cases, against 45 of 247 for the rest, though feeds from one publisher tend to lapse together.

A forecast for 1 April 2027

Of the 574 active Japanese feeds on 16 October 2026, 292 (51%) were set to expire on 31 March 2027. We recorded a forecast before that date: if the last three years' rates of 14.3% to 20.2% hold, about 42 to 59 of those 292 feeds, with a central estimate of about 47, will have no valid file on 1 April 2027. Further forecasts, also committed in advance, allow for the rising trend and for publishers lapsing together. A back-test showed that predicting each year from the year before has run low, because the share has mostly risen. We will re-run the census around 8 April 2027 and score every forecast, including any that fail.

Pre-registered forecasts for 2027-04-01

ModelFeeds at riskExpected lapsesRange
Last three years' rates (simple)292~4742-59 (from the 14.3-20.2% rates)
2026 rates by group (prior lapsers 40.0%, others 18.2%)29461.645-78 (rough 90%, widened by publisher)
2024-2026 rates by group29452.737-69 (rough 90%, widened by publisher)
Straight-line trend, 2022-202629469.051-87 (rough 90%, widened by publisher)
The simple forecast uses the 2026-10-16 count of 292 feeds; the group and trend forecasts use the 2026-10-22 list of 294. Source: gtfs-jp-lapses/README.md and forecast/2027-04-01/.

What it means

Most expired timetables in the catalog are expired at the source, so the fix lies with publishers renewing on time, especially around predictable dates such as Japan's fiscal year. A small set of stale copies is the catalog's to fix, and those cost riders real services that exist. Apps that read catalogs should check feed end dates and warn riders rather than show an empty schedule.

More research