Where the railway makes trains slow down

Network Rail publishes every stretch of track where trains must run slower than normal. Reading it teaches you something the word “temporary” rather hides: most of them have no end date, and one has been in force since 2004.

When a train is booked to take fifty-two minutes and reliably takes fifty-eight, the reason is often not the train, the crew or the signalling. It is a few hundred metres of track with a lower speed limit than the timetable was written for.

These are temporary speed restrictions, and they are published weekly in the operating notice. We list the ones currently in force on the speed restrictions page.

What a restriction actually says

Each one names a stretch of line between two places, which road it applies to, a speed for passenger trains and another for freight, why it is there, and the dates it runs between. A real one, taken from the notice:

Eastleigh South Jn to Eastleigh South Jn · Down Portsmouth · 74.0000 to 74.0875 miles · 70 mph passenger, 30 mph freight · from 2 December 2025 · Track — cyclic top

Two details there repay attention. The line name matters: “Down Portsmouth” is one specific road, and a restriction on it is not a restriction on the line beside it. And the two speeds are usually different, because a heavy freight loads the track very differently from a lightweight multiple unit over the same defect.

The reasons are the operating notice's own words and we reproduce them unchanged: “Track — cyclic top”, “Safety — level crossing sighting”, “Safety — signal sighting/braking”, “Condition of earthworks”. Cyclic top is a repeating dip in the rail that a train can start to bounce in sympathy with. Earthworks means the ground the railway sits on rather than the railway itself.

“Temporary” is the industry's word, not a promise

Here is the measurement that changed how we present the whole list. Counting the restrictions in force in September 2026, only 78 of 280 stated when they would lift. The rest have no end date at all. The exact figures move as restrictions are imposed and removed, but the proportion has been steady every time we have looked: most of the list does not say.

And they are not all recent. The oldest in force runs between Shell Junction and Eastgate Mount level crossing, at 10 mph for both passenger and freight, with the reason given only as “Track — other”. It has been in place since 27 March 2004 — twenty-two years. Ten more have run since before 2020.

This is not a scandal and it is not a filing error. Some restrictions are cheap to live with and expensive to remove — a permanent speed reduction through a curve with sighting problems may simply be the settled answer. But it does mean that reading the list as “things about to be fixed” would be quite wrong, and it is why our page states the number without an end date rather than leaving the blanks to look like missing data.

The date that means “never”

The feed has a convention we nearly published as fact. Restrictions with no planned end carry an end date of 1 January 4000.

Kept as a date, that renders as an answer. Our first working version of the page would have told you a restriction “lifts on 1 Jan 4000”, with about 712,000 days remaining. On the list as it stood that day it was 78% of entries — 193 of 249 — so most of the page would have said it.

We now map anything beyond a sane horizon to “no end date”, which is the absence that is really there. The check is a horizon rather than an exact match on that one value, because a feed that switched to 9999-12-31 tomorrow should not quietly bring the problem back.

It is a shape of bug worth knowing generally: a placeholder that happens to be a valid value of its type. A date of 4000, a livery code meaning “unset”, a zero that means “stop” rather than “no data”. Each one reads as real until someone checks.

Why the mileages are shown as published

Each restriction carries a mileage along its route — 74.0000 to 74.0875 in the example above. We show those exactly as the notice states them, and we do not convert them into anything.

The reason is that the notice's mileage frame and the one used elsewhere on this site are not guaranteed to be the same reference. Rather than pretend they are, restrictions are matched to a planned route geometrically — by working out where on the ground the named places are and which stretch of track lies between them. When you plan a route across one, that is how it gets attached to the leg.

The same caution appears in how the railway names its own track: a mileage only means something once you know which measuring frame it belongs to, and there is more than one.

Two things the raw notice will do to you

The published list repeats itself. Every time we have counted, some references appear more than once — up to three times — with the repeats identical, most likely because one stretch is listed under more than one route. In September 2026 it was eleven duplicates in a list of 291, leaving 280 distinct restrictions. Any straightforward count is therefore slightly too high until you deduplicate.

And a speed of zero is a real value, not a missing one. It means stop. A reader that treats zero as “no data” — which is the natural thing to write — will silently drop the most severe restrictions in the file.

What the list will not tell you

Roughly a third of restrictions can be placed on a map. The rest name locations we cannot resolve to coordinates — junctions and loops that do not appear in the location reference we hold. Those are still listed, because a restriction we cannot draw is still in force, and leaving it out would under-report the railway rather than merely under-draw it.

It also will not tell you how much time any of them costs. A restriction's effect depends on the line speed it replaces, how long the stretch is, and whether the timetable was written with it already in place — a long-standing one is usually absorbed into the booked times, so it causes no delay at all while still slowing the train.

That is the part worth carrying away. A speed restriction and a late train are related, but far less directly than the word suggests, and how a delay is measured covers the other end of that story.