How accurate is the data shown on RealTimeTrains?

[The Planner]

If it is guessing in that way when it is late it should really add the SRTs and minus the pathing and performance allowances.

 

[Tom]

RTT is not so good at predicting the future.

Predicting the future is difficult. To do it well, you need what is essence artificial intelligence which is able to understand the human brain and determine what regulation decisions are likely to be made. This is interesting reading on those lines.

The prediction model used on the present public version of RTT is relatively primitive: it has some mathematical weighting, looks at run times of previous trains to some degree and the lateness of the existing train. The evidence suggests that the other system that does predictions does much the same in the vast majority of cases. With that in mind:

It was some 15 minutes late, so it lost its sequence and meant it would conflict with the CLC stopper from Oxford Road.
Turning this over in my head, I decided to let the EMT go and get the following Airport-Liverpool 319 via Chat Moss instead, which was pretty much to time.
While travelling I kept track of the CLC trains, and predictably the stopper was let out at Oxford Road on time and duly held up the EMT service all the way down the line to Edge Hill.
In the end the 319 overtook the EMT and arrived 15 minutes ahead of it at Lime St.
Meanwhile the RTT feed continually showed an over-optimistic arrival for the EMT service, not correcting for the delay behind the stopper.

RTT will look, in some very limited areas, at trains ahead of the existing one where I have modelled the track (pointwork, lines, etc) so it can attempt to 'follow on' and attempt to second guess any regulation decisions, such as looping. Needless to say, this isn't one of them.

So it falls back to its secondary model of effectively following the path, and attempting to determine anything in front of it. It can't do this beyond any known junction (the mileage engine allows for this, but the mileage engine doesn't have any data as to which route is accessible from which direction, hence odd circular routes sometimes) and there are a number of cases where it can't do it at all as there are other geography related reasons and processing limits to ensure backlogs don't build up.

In this case it goes to a third model of just following the path, but applying known variables to it (including considering running times) which sounds like what happened here. This model is fairly optimistic in that it expects that a late running train would likely be routed in a way such that it can get ahead of stopping train that is on time, which is normally the case in my experience, but you can't get everything right.

All three models take account of sectional running times and any allowances as provided in the schedule.

In any case in instances like this, feedback is appreciated so it can be reviewed and be used to improve the system in the future. In the last 12 months I think I can safely say that the comments via the feedback channel on the prediction side of things has been a grand total of zero. As I said elsewhere, we don't read forum posts all the time (I was passed this one by louis97).

On a side note, it's worth noting that many industry systems have huge budgets purely for system development (hundreds of thousands of pounds, in some cases millions). I have a development budget of about, er, nil and yet, most of the time, come very close to reality. Even the hosting budget for many systems is huge - whereas at the moment I'm struggling to find the money to go for a ~£10,000 upgrade on new servers and a new hosting location so that RTT can be improved in the future. Difficult times.

 

[LNW-GW Joint]

On a side note, it's worth noting that many industry systems have huge budgets purely for system development (hundreds of thousands of pounds, in some cases millions). I have a development budget of about, er, nil and yet, most of the time, come very close to reality. Even the hosting budget for many systems is huge - whereas at the moment I'm struggling to find the money to go for a ~£10,000 upgrade on new servers and a new hosting location so that RTT can be improved in the future. Difficult times.

Thanks for posting the detail Tom. I wasn't intending to be critical at all.
It must be a bottomless pit of mathematical modelling to allow for all the possibilities.
I imagine the fancy TMS systems that NR are developing with the big boys will one day model the whole network for the signallers, but in the meantime it's great to have RTT as a window on the rail universe while travelling.
It also never misses a beat, or suffers from endless upgrades.
As an old programmer myself, I'm really impressed with it.
I hope you can keep it going.