The main three things are:

1) very complicated and drawn out design/approval processes with many expensive consultants involved. on network rail you would not believe the amount of studies they have to do, including testing the light levels everywhere in a carpark to make sure there are no dark spots, amongst a billion other small things. there has been endless regulation forced on the railways/tfl, which on one hand is good for safety but on the other has meant even a tiny project needs a trillion risk assessments done, which typically get outsourced to specialist consultants, who have a massive incentive to find problems so they can retest.

2) lack of competition at the top level contractor layer. the procurement rules require really only huge companies can bid, which means they can act in a semi-cartel way. this is probably difficult to solve given signalling stuff but in more basic construction projects it feels like not much competition is going on (not suggesting there is a cartel here, fwiw, just feels a bit like it!)

3) on the tube the access is extremely restrictive. you tend not to close the line for extended periods of time (though this is changing recently as i think people are realising the cost of keeping the line isn't worth it vs doing a week closure), which means that you have to get all your equipment down to the tunnel (which might taken an hour), do 2-3 hours of work, then remove all the equipment, in 4-5 hour blocks each night between midnight and 5amish, which means you are paying people often "night working" wages for working really 2 hour shifts, instead of paying them normal wages for working through the day for 8 hour shifts. This probably increases the cost of labour by 5-10x.

That's the price of making complex, real-world hardware systems fail-safe. By fail-safe, I don't mean 'never fail' but, more literally, fail in a way that remains safe for all other systems interfacing with the failing one.

Proving that it fails safely requires a mountain of effort and documentation, but it's required because failure doesn't just lead to some speculative CPU bugs leaking a password or to a browser crash, it leads to many people dying.

To add to your point, the accident in Greece this summer that killed 50+ kids was attributed to human error that could've been avoided if they had upgraded their signalling system like it was promised decades ago.

I get that, but still, it is not like we don't have experience with making fail-safe things that we can draw from. And once you get a station + track fitted with the new system you just copy/paste; so I can understand a higher cost in the beginning for the proof of concept, but after that everything should be very cheap.

"you just copy/paste"

Aren't we talking about physical signalling kit - no copying and pasting? 272 stations and ~400km of track - a lot of which is pretty difficult to access?

He obviously meant copy and pasting the design. What did you think he meant?

I have no idea - that's why I was asking. I've been responsible for installing systems that had the same "design" into multiple physical locations - hundreds of physical devices at each location.

Having a common design does help but it doesn't help that much as most of the problems are with the physical realities of each location - and what I did was much simpler and less critical than signalling.

Ahhhh! "just" - the magic word!

It is not ever 'just'. This isn't copying code. This isn't building identical silicon for multiple end users. This is more like trying to add additional threads to something that is not embarrassingly parallel. What you're saying is like saying 'why can't you just make all of my computer programs run on all 32 of my CPU cores?'.

And even if you minimise the increasing complexity added by expanding the system, waterfall-style design (which is absolutely the most appropriate approach when dealing with these sorts of projects) front-end loads a frankly Herculean amount of risk management into the process because you want to make sure that's all absolutely square and true before you progress to implementation.

Lastly, it's worth noting that with systems this complex, a lot of the detail is bespoke because the number of complicating input variables the system is exposed to necessarily requires a relatively bespoke solution. Yes, the idea of e.g. ETCS being ETCS is going to hold, but how that works in reality, how it ties in to all of the other legacy systems, etc. will mean that you can't for example just take ETCS-equipped train A and and drop it into ETCS-equipped network B.

I was thinking that only the track sensors and traffic lights along the way need to be upgraded and the drivers would just drive trains as usual and they would be sequenced on a shorter distance/time between trains. I forgot that trains also have to be connected to the new system.

That’s less than £10 per rider of the line. Piccadilly line ridership is over 200 million trips per year.

This is a line that has 53 stations. For comparison that’s 20 more stations than the longest and highest volume L line in Chicago.

I wonder how this cost benefit compares to the $9 billion project to widen Houston highways?

What does £10 per rider of the line mean? How does that help us look at things?

I used to commute 5 times a week (often more) on Piccadilly. So that’s £5,200 for my ridership for the year? From that perspective seems rather expensive…

It’s disappointing they ditched the signal upgrades. It will bite everyone soon enough. Some of these systems date back over a hundred years, and are all incredibly bespoke.

Re the missed stations, I feel like they missed an opportunity with the Elizabeth Line line to shift more Heathrow traffic that way, giving some breather to Piccadilly during signal upgrades.

The average American spends $5,000 a year on the total cost car ownership, completely separate from public spending on roads and infrastructure, so I think that signaling cost is reasonable enough, especially since signaling will last far longer than the typical lifespan of a personal vehicle.

£5200 for your annual ridership but then divide that by the lifespan of the signaling. Let’s say it lasts 30 years, so that’s €173, (but of course, at the end of the 30 years £173 is worth a lot less).

How much an American spends on a car every year seems not that relevant to how much a British tube line signal upgrade costs. Why not compare it to house prices in London, and it'll seem even cheaper?

I compare it to automobile infrastructure because public transit infrastructure projects sems to be the only time anyone complains about high infrastructure costs.

The only reason I compare it to America is because I don't actually know the figures for the UK, but I expect that car owners would spend more there due to higher cost of petrol, higher vehicle taxes, etc.

Cars are unreasonably useful, though. If you're going anywhere to work / school / moving house / going on holiday / dropping kids at a friend's house, they work for a million uses.

That's why I don't think they compare well: many people who use public transport, even the Tube or similar, still need a car, at least sometimes. So it's not an alternative, and it's not equivalent value.

So probably better to pitch it not as £10 per rider (~200m annual riders) but £0.33/trip levy for the next 30 years.

What do you mean by missed opportunity? To not run it there more frequently? Can't they still do that? The timetable has seen some updates already.

I’m routinely surprised by how much things like this cost.

I think the answer is that solving this problem takes a teak of some size and skill, and paying them while making a hefty profit is expensive.

It’s compliance and bureaucracy that keeps competition out.

I think it has more to do with lack of state capacity. The agencies responsible for building this don't have anyone in-house anymore who can actually plan and carry out the job, instead relying on cascading layers of consultants and sub-consultants who manage contractors and sub-contractors.

Corruption