iiinova — where your digital self keeps evolving

Uday Schultz @a320lga.bsky.social · 05/10/2026

minimum time between when one train begins to leave a track and the next train fully berths in the track

Uday Schultz @a320lga.bsky.social · 05/10/2026

Yes! At a very high level, max turn time at a terminal is basically: (number of trains you can fit in your turning area * scheduled headway of service) - turnaround track reoccupancy time

140

Uday Schultz @a320lga.bsky.social · 05/10/2026

I might change the 08 to a 12 to even out the gap in service. this is a simplification of course but you get the idea

050

Uday Schultz @a320lga.bsky.social · 05/10/2026

additionally, though the subway is a schedule managed network, dispatchers have the power to (and are encouraged to) do some headway management around anticipated cancellations. so, for example, if I have trains on an 8 min headway and I know i'm going to lose the :16 departure

150

Uday Schultz @a320lga.bsky.social · 05/10/2026

...so e.g. a 1208 arrival that was going to make a 1220 departure might make a 1225 or 1230 if it arrives, say, 13 mins late.

140

Uday Schultz @a320lga.bsky.social · 05/10/2026

as for what makes NYC distinctive: a lot of operators are bound to their equipment blocking. the trip that comes in late goes out late as whatever its next scheduled trip is, and that's that. NYC is different: you can (and sometimes are required to) adjust inbound-outbound pairings at terminals

160

Uday Schultz @a320lga.bsky.social · 05/10/2026

eh, you can do creative things when expressing high-frequency services. give two trains skips back to back, have the first trip skip stops between A and B and then the second skip between B and C, etc

140

Uday Schultz @a320lga.bsky.social · 05/10/2026

that's what it would *seem* yes. unfortunately no turn time data in the public schedule data, though for some agencies you can use block_id to get at this info

150

Uday Schultz @a320lga.bsky.social · 05/10/2026

yeah if they're scheduled to do a return trip things aren't terrible. it's crews on their final run, crews that have lunch, and crews that are scheduled to interline to some nonrevenue activity that tend to cause agita in these situations

151

Uday Schultz @a320lga.bsky.social · 05/10/2026

or worry about what damage you're going to do crew schedules by turning a train away from its scheduled destination

140

Uday Schultz @a320lga.bsky.social · 05/10/2026

yeah in a truly high frequency metro context running express often is a bit easier than turning, esp if you don't have good places to discharge trains when turning

140

Uday Schultz @a320lga.bsky.social · 05/10/2026

yep! a lot of it has to do with infrastructure and operating rules. NYC for example is nearly entirely reliant on expressing for this purpose (tho NYC also has unusual terminal dispatching rules)

280

Uday Schultz @a320lga.bsky.social · 05/10/2026

yes to all of this. I have to imagine that the fact that the S Bahn is graded on a cancellation metric that treats anything more than 1 headway late as cancelled helps strengthen this incentive here too

1100

Uday Schultz @a320lga.bsky.social · 05/10/2026

and if you have a decent traffic management system (or dispatchers who are trained in re-dispatching) it's relatively trivial to game some of this stuff out well in advance

130

Uday Schultz @a320lga.bsky.social · 05/10/2026

I guess I should say for the record that I am (generally) very in favor of active service management. I'd say the median rail operator tends to under-do it.

170

Uday Schultz @a320lga.bsky.social · 05/10/2026

yes, or re-book on the following train.

120

Uday Schultz @a320lga.bsky.social · 05/10/2026

got it, so (for example) S1 Frohnau and S1 Oranienburg are internally two different things

190

Uday Schultz @a320lga.bsky.social · 05/10/2026

I am hardly an expert so others plz correct me, but that article would suggest so! As Nilo said in the other thread branch, this seems to be an artefact of service density

010

Uday Schultz @a320lga.bsky.social · 05/10/2026

heh, yep. games like these become exponentially more difficult to play as stopping pattern complexity increases and frequency decreases.

150

Uday Schultz @a320lga.bsky.social · 05/10/2026

"these english speakers can have a little german rail operations research, as a treat"

2130

Uday Schultz @a320lga.bsky.social · 05/10/2026

We could add time to the long trips to protect the shorts from lateness, but in that case there was a way to move the long trips to a different track at that location to eliminate the interdependency. So we did that, and eliminated the critical point

0110

Uday Schultz @a320lga.bsky.social · 05/10/2026

I worked on a project once where we were asked to decompose a high departing lateness rate from a mid-route terminal. The cause of the issue was tight slots with trips from the further terminal that would lead to departing delays for the mid-route variant with even minor lateness.

190

Uday Schultz @a320lga.bsky.social · 05/10/2026

what you want to avoid are tight interdependencies ("critical points") between trains that are also not well-protected by running margin from delays inbound to the point. you can solve that by reducing the pinch of the interdependency (engineering or service simplification), or adding time.

290

Uday Schultz @a320lga.bsky.social · 05/10/2026

would second both the above observations. this is why you pretty consistently see lines with mixed-speed traffic having much more running time margin.

2100

Uday Schultz @a320lga.bsky.social · 05/10/2026

Yes, same goes for en-route margin. You should be engineering/ops-reforming your way out of as much variability and interdependency as possible, then applying margin.

060

Uday Schultz @a320lga.bsky.social · 04/10/2026

The (obvious, if helpful) takeaway: designing timetables for reliability, and then designing your dispatching regime for that timetable (and your infrastructure) is quite effective.

2350

Uday Schultz @a320lga.bsky.social · 04/10/2026

The differences in cancellation probability visible at the line/terminal further underscore this point, suggest the likely moderating structure of these actions: differences in terminals' scheduled turn times. Longer scheduled turns = more capacity to absorb delay.

2220

Uday Schultz @a320lga.bsky.social · 04/10/2026

This appears to reflect a dispatching regime that with clear rules/procedures to restore service to schedule during/following disruptions.

1260

Uday Schultz @a320lga.bsky.social · 04/10/2026

It's remarkable how predictable these turns are. Across all lines, trips passing their midpoint of their routes close to schedule will likely complete their trips. But once trains fall more than 4 mins behind schedule, the probability of partial cancellation shoots up.

1270

Uday Schultz @a320lga.bsky.social · 04/10/2026

On to the second point (short turns). You may have noticed in the previous punctuality graph that: -Cancellations increase towards the end of the line -That increase in cancellations is correlated with a reduction in lateness rates. This is a pretty standard signature of short-turning for lateness.

1270

Uday Schultz @a320lga.bsky.social · 04/10/2026

Despite the nontrivial lateness rates shown above, even the busiest sections of the network seem to have high in-sequence rates, suggesting that: - Timetable designs network-wide are optimized to preserve train order - Dispatchers are willing to delay trains a bit to keep them in order Structure!

A chart showing the out-of-sequence trip rates by stop for berlin's S9 line. The highest value of out of dequence trains is around 8%.

A chart showing the out-of-sequence trip rates by stop for berlin's S2 line. The highest value of out of dequence trains is around 8%.

1340

Uday Schultz @a320lga.bsky.social · 04/10/2026

I spend my working life fighting with the performance intricacies of a highly interlined rail network, so naturally I was curious to understand how the S Bahn handles sequencing issues at merges.

1360

Uday Schultz @a320lga.bsky.social · 04/10/2026

These buffers have a clear impact on reliability. Here's the en-route punctuality data I derived from the dataset above for the S9. Note the steep change in lateness levels around Treptower Park, esp in the BER direction.

this graph shows punctuality rates (lateness + cancellations) by stop for Berlin's s9. two notable features: cancellation rates tend to increase towards terminals, and lateness rates sharply drop around the treptower park buffers shown previously.

2300

Uday Schultz @a320lga.bsky.social · 04/10/2026

On the first point: the S Bahn runs a patterned 20/10/5 min headway timetable throughout the day on all lines. The consistency of this structure appears to facilitate easy/consistent adjustments of slotting and buffers in the timetable, as can be seen here at Treptower Park:

both of these stringlines show the S9's timetable. key element: extra time placed at Treptower Park for slotting and (presumably?) delay recovery

2380

Uday Schultz @a320lga.bsky.social · 04/10/2026

The overall "vibe" you get is that: - They have an extremely structured service management regime at the timetabling and dispatch level - They rely quite heavily on short turns to maintain reliability

2441

Uday Schultz @a320lga.bsky.social · 04/10/2026

This wonderful dataset has allowed me to get a few glimpes under the hood of the Berlin S-Bahn. And it's fascinating! huggingface.co/datasets/pie...

huggingface.co

piebro/deutsche-bahn-data · Datasets at Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

1401

Uday Schultz @a320lga.bsky.social · 04/10/2026

One of the questions that has long flourished in my head is how, tactically and materially, transit networks abroad actually manage their service to produce better reliability outcomes than we're used to seeing on this side of the Atlantic.

4979

Uday Schultz @a320lga.bsky.social · 29/08/2026

Lovely illustration of how important terminal dispatching is for transit. This graph shows the progression of vehicles on the 49 today; I've circled this bunch coming out of the terminal. Had someone told one of these bunched buses to run express for a while, you would have cleaned this up!

14211

Uday Schultz @a320lga.bsky.social · 29/08/2026

On the other hand, though, it's genuinely unclear to me how much of an impact reform plans would have on *highway* volumes given the ever-shrinking acerage of industrial land in the city. This is the "after" map of projected garbage truck traffic post-CWZ implementation.

160

Uday Schultz @a320lga.bsky.social · 29/08/2026

I am ambivalent about planning's capacity to solve this. On the one hand, there are probably both pricing and regulatory measures that the city could implement that would rationalize intra-city truck traffic. The estimates for commercial waste zone rollout are like a 50% reduction in traffic!

180

Uday Schultz @a320lga.bsky.social · 29/08/2026

(Trucks then were a little less than 10% of traffic on the cantilever then) Given the changes in NYC's industrial geography since then, and the general increase in first/last mile frieght activity, my intuition would be that both the truck/commercial vehicle % and the NYC % have increased

181

Uday Schultz @a320lga.bsky.social · 29/08/2026

Replacing the central BQE with something for trucks/buses only has always felt to me like the most potentially interesting removal option to analyze. The data I have are old, but something like 70% of BQE truck traffic was intra-NYC in the early/mid-2010s

2141

Reposted by Uday Schultz

Laurie! Live! @laurie-merrell.bsky.social · 09/08/2026

I have also written a couple of blog posts about working with transit data, here's the first one -- a general intro to some of the data in the space lauriemerrell.com/posts/transi...

lauriemerrell.com

So you want to do a public transit data project, part 1: Working with common sources

A short guide to common types of transit data and where to find them.

192

Uday Schultz @a320lga.bsky.social · 09/08/2026

Yeah — my expectations re what I’d find when I looked at their blocking data were similar to yours so I’m fascinated by these results. Seems to be true of their tram networks as well

050

Uday Schultz @a320lga.bsky.social · 09/08/2026

in retrospect, would have been a great project for claude or similar

160

Uday Schultz @a320lga.bsky.social · 09/08/2026

i spent an afternoon a few months back downloading euro gtfs feeds until i found ones with block ids 🥲

260