Dodgy train times/performance recording?

[rorbla]

Hi all,

Long time lurker, first time poster.

I caught the 15:52 from Paddington to Oxford on Saturday. It was delayed at departure, and slightly more during the journey. Realtime trains is showing that the train arrived into Oxford at 16:57.

However, the train absolutely did not arrive into the platform until 16:58, and I took a photo of the platform information screen showing 16:58 whilst the train was still pulling in. This would be a 15 min delay and thus subject to delay repay.

I believe Realtime Trains is the publicly available version of the data used by Network Rail and the TOCs. Is this correct?

I presume NR collect this data automatically from their signalling systems, rather than the TOC report this back to NR?

If so, is there a dodgy practice taking place here where NR and/or TOCs are deliberately fudging performance to minimise delay repay payments?

EDIT: And, conveniently, Oxford has a 3 min dwell time (so departed at 17:00), versus all subsequent stops which have a 1 min dwell time.

All subsequent stops were subject to a 15-16 mins delay too (versus an incorrect 14 mins delay to Oxford).

 

[The Planner]

Hi all,

Long time lurker, first time poster.

I caught the 15:52 from Paddington to Oxford on Saturday. It was delayed at departure, and slightly more during the journey. Realtime trains is showing that the train arrived into Oxford at 16:57.

However, the train absolutely did not arrive into the platform until 16:58, and I took a photo of the platform information screen showing 16:58 whilst the train was still pulling in. This would be a 15 min delay and thus subject to delay repay.

I believe Realtime Trains is the publicly available version of the data used by Network Rail and the TOCs. Is this correct?

I presume NR collect this data automatically from their signalling systems, rather than the TOC report this back to NR?

If so, is there a dodgy practice taking place here where NR and/or TOCs are deliberately fudging performance to minimise delay repay payments?

EDIT: And, conveniently, Oxford has a 3 min dwell time (so departed at 17:00), versus all subsequent stops which have a 1 min dwell time.

All subsequent stops were subject to a 15-16 mins delay too (versus an incorrect 14 mins delay to Oxford).

No, there is no conspiracy theory. The berth offset may be slightly out. Not sure what the dwell has to do it with it either, lots of major stations have longer dwells due to the amount of people using them.

 

[rorbla]

I’m not suggesting a conspiracy theory. I work in a similar, regulated field and policies such as delay repay can result in perverse behaviours; it’s a natural response where regulation allows it. Ultimately, it depends on the nature of the system; where responsibilities lie and how the data is collected.

My point with Oxford having a long dwell vs all other stations served on that journey was suggesting that the time spent rolling into the platform (still moving) potentially being allocated to the dwell time vs ‘approach’ time. Perhaps this is essentially the berth offset?

Perhaps this is related to the Botley Road Bridge/Oxford Station works, which have been going on for 3+years? On one hand completely valid, on another hand convenient after 3+ years?

My Delay Repay claim was rejected, which I have appealed. I’m sure many others on that train will submit claims, but not reject the appeal. Ultimately, given the policy of delay repay, someone (GWR or NR or both) should be liable for more compensation payments than they’re likely to pay in this instance. Furthermore, this could be a frequent occurrence.

I don’t really mind the money. I do mind the extortionate ticket price from London to Oxford for an Off Peak journey, the regular delays on GWR’s routes (not always GWR’s fault), Saturday’s train making me late for a meeting, and the Botley Road Bridge works being dragged out by NR. It’d be adding insult to injury if GWR or NR are then arbitrarily applying an incorrect offset (potentially to minimise their delay repay liabilities/artificially improve performance). Other people may really need that money though.

 

[Vexed]

Essentially RTT takes raw berth signalling data from Network Rail and uses its own dataset of offsets to work out arrival and departure times. The rest of the industry uses Network Rail's SMART dataset for offsets. In detailed mode, RTT puts the arrival at 16:57:30 (note RTT rounds to the nearest 15 seconds).

A system called TRUST will offset the signalling movement data with the SMART data to record the official delay, which works in whole minutes and is always truncated not rounded. Using the berth step at 16:56:50 and adding the +66 second offset this gives an arrival at 16:57:56 - truncated down to 16:57. Just 4 seconds off a 15 minute delay. Ultimately it's the difference in driving styles between individual drivers and also if you were approaching slowly on restrictive aspects the offset won't represent reality very well.

Some newer trains are able to record the door release time but I don't think this is publicly available.

I would expand the SMART and TRUST acronyms but what they stand for doesn't really mean anything in isolation.

 

[Tom]

note RTT rounds to the nearest 15 seconds

Not quite, it always rounds down because of cases like this. RTT actually puts the arrival at 16:57:44 internally.

 

[rorbla]

Thanks again for this Vexed - really helpful to understand and discuss.

To confirm my understanding; so, the time shown on RTT is the point at which the train passes the signal into the station ‘block’/berth?

NR then apply a pre-specified offset to this to allow for the driver to pull into the platform, stop and open the doors?

If so, motivations aside, surely that latter part is subject to a range of variables that could affect whether it’s correct or not?

Does anyone know where the signal is on the ‘up’ approach into Oxford (is it before Botley Rd Bridge where there are major works underway and trains approach relatively slowly?)?

Also, is the time shown on RTT the time at which the train first passes the signal, or is it when the whole train has cleared it?

 

[Vexed]

To confirm my understanding; so, the time shown on RTT is the point at which the train passes the signal into the station ‘block’/berth?

NR then apply a pre-specified offset to this to allow for the driver to pull into the platform, stop and open the doors?

NR's Train Describer system gives berth-level locations, with modern signalling systems each berth is usually the same as a signalling block.

RTT do apply offsets - it's not the time passing the signal. That's the time RTT receive then they offset it themselves.

Network Rail offset from the same berth step (each movement in the Train Describer system aka passing a signal is called a step) but use a different data set and go to quarter minutes instead of whole minutes.

Also, is the time shown on RTT the time at which the train first passes the signal, or is it when the whole train has cleared it?

As above RTT's time is offset for the actual arrival at the location.

The trigger for the berth step that NR and RTT offset from can vary depending on the signalling system. It can be the front of the train passing the signal, or on plain line if the overlap is in the same track circuit as the approach to the signal it might step up to 183 meters (200 yards) beyond the signal. But this should be accounted for in offsets.

 

[Tom]

If so, motivations aside, surely that latter part is subject to a range of variables that could affect whether it’s correct or not?

Correct, but multiple trains are measured in a range of different conditions which leads to a statistical judgement of the 'average' train. For NR it's 6 services, for RTT it's 4 but under more specified operating conditions. We also have a monitoring process that validates offsets.

Does anyone know where the signal is on the ‘up’ approach into Oxford (is it before Botley Rd Bridge where there are major works underway and trains approach relatively slowly?)?

It is a long distance before that bridge assuming you are referring to the one immediately south of the platforms, but that likely will be causing an impact if there is a temporary speed restriction in place (but the permanent restriction appears to only be 30 anyway into 4).

Also, is the time shown on RTT the time at which the train first passes the signal, or is it when the whole train has cleared it?

The time shown is the calculated offset after the 'step' report at the signal, offsets have been explained above. In this case, the trigger point looks like it'll be 180 metres beyond the signal.

 

[rorbla]

Interesting.

I have always assumed the railway was capable of reporting, more or less to the second, the actual time at which a train arrives into/departs from a station.

However, it appears that arrival and departure times are modelled.

I could have been on a train that is at the lower end of the performance distribution, with an actual time of arrival later than what is an ‘estimated’ time of arrival (based on average performance). In this case, this approach seems to have resulted in an estimate below the 15 min delay threshold vs the actual being above it.

Obviously, this is a highly marginal case and an unlikely situation so I doubt there is an ulterior motive to this system, but I do wonder how much delay repay compensation has been saved by TOCs and NR as a result of this.

Also, as a slight aside, with ETCS on the way and with the higher levels seeing no trackside equipment and moving blocks, how would this be calculated? Would the blocks/berths be retained as is, but in digital form, and used in the same way?

== Doublepost prevention - post automatically merged: 26 Aug 2025 ==

Tom, you mention the offsets are validated/monitored.

How is this done out of interest? Are there other data sources to triangulate to?

Would the speed restriction be factored in? Would the likely poorer performance as a result of the bridge and station works be factored in?

 

[Tom]

There needs to be a national solution to do it equitably really using the same process everywhere possible to remove the ease of disputes from disagreeing datasets.

Also, as a slight aside, with ETCS on the way and with the higher levels seeing no trackside equipment and moving blocks, how would this be calculated? Would the blocks/berths be retained as is, but in digital form, and used in the same way?

Berths are still a concept used in ETCS areas but they may or may not be more dense. In theory if the same model is used then the error tolerance will be reduced but still not necessarily precise.

But there is a broader question here - what constitutes an arrival/departure time? Is it the time the wheels stop/start or when the doors are released/closed? Those answers will be different from different angles. In airline compensation, the key time on arrival is when passengers are able to first disembark from the aircraft, rather than brakes on or doors open.

Tom, you mention the offsets are validated/monitored.

There is an audit process for the mainline (you should lookup the Performance Delay Accuracy Code) as there is for RTT.

How is this done out of interest? Are there other data sources to triangulate to?

There is not a huge amount of other data sources out there to reference against, but there are a few and we will check against the industry's official view as well as a self referential integrity check (i.e. have dwells suddenly got a lot higher for these trains, etc)?

Would the speed restriction be factored in? Would the likely poorer performance as a result of the bridge and station works be factored in?

Potentially, but I can't answer that question. I suspect the answer is no for SMART - it looks like the offsets were last changed in 2020 (this is available openly in the SMART dataset). For RTT it was more recent but for various reasons I won't share the exact date.

 

[rorbla]

This is great, thanks.

Glad you mentioned dwell time, as that aligns with my thoughts in my initial post. Extended dwell times could mean the offset is too generous. I’d be interested to see what dwell times look like at Oxford over the years, pre-post Botley Road Bridge works. Several potential confounders, but these could likely be controlled for, assuming the data is available.

In fact, the above could probably be used to validate/identify outlier offsets automatically.

 

[The Planner]

This is great, thanks.

Glad you mentioned dwell time, as that aligns with my thoughts in my initial post. Extended dwell times could mean the offset is too generous. I’d be interested to see what dwell times look like at Oxford over the years, pre-post Botley Road Bridge works. Several potential confounders, but these could likely be controlled for, assuming the data is available.

In fact, the above could probably be used to validate/identify outlier offsets automatically.

Your train on Saturday presumably was 1W51 Paddington to Moreton in Marsh. The schedule had a 2½ minute dwell at Oxford between 1642½ and 1645 (the mandated minimum for a Class 800 at Oxford is 2 and is standard for 800s at most major stations on the Western such as Reading, Swindon, Bath, Exeter and Plymouth), so it was extended for 30 seconds for reasons GWR felt necessary.