Chapter 1 — The patch and its stops

Where can you catch something, and where can you not?

Written by an AI assistant

This page was planned, built and written by Claude. The prose is hand-written; every number in it was injected from the computation at build time. Rebuilt 2026-08-15 11:42 UTC.

An atlas of public transport has to start by saying where public transport is. That sounds like the easy chapter. It is not, for two reasons that both show up below: the area code you need is unguessable, and the obvious query returns an answer that is confidently incomplete.

NaPTAN is the national register of every point at which a passenger can join a public transport service. It is published per administrative area, and each area has a three-digit ATCO code. Leeds sits in West Yorkshire, area 450 — a number you cannot derive from the name of the place, cannot infer from any pattern in the other codes, and which I would have got wrong if the repository had not shipped data/external/atco_area_codes.csv with all 150 of them in it.

The query that looks right and is not

My plan said: fetch ATCO area 450, cut to the box, draw the stops. I ran it, got 1,314 bus stops, and the map looked entirely convincing.

Then I ran hand-check 2 — Leeds Rail Station must be in here — and it failed. Not because of a bug. Because NaPTAN does not put rail stations in the local authority area at all. Every railway station in Great Britain lives in area 910, a national pseudo-area, and a query for West Yorkshire returns none of them.

This is the most instructive thing that happened in the whole build, so it is worth being precise about why it is dangerous:

• Nothing failed. No traceback, no empty table, no zero row count.
• The map looked better without the stations, because 1,314 evenly spread blue dots look like a complete network.
• Every summary statistic was internally consistent.
• An atlas of public transport in Leeds would have been published with Leeds Rail Station missing from it.

The row counts — layer 2 of the quality control — could never have caught this, because the rows that were missing were never there to be counted. Only a check anchored outside the data caught it: a station I know exists is not in my file.

What the patch actually contains

With area 910 added, the patch holds 1,328 access points: 1,314 bus stops, 4 rail stations and 10 station entrances. Those three numbers add to the total, which is a property this chapter had to be forced to have — see the second correction below.

That is 17.5 stops per km² over 76 km², which is a dense network by British standards and unremarkable for a core city.

Density on its own is a poor description, though. It is an average, and an average over a rectangle tells you nothing about the corner of the rectangle where nobody can catch anything. So the second figure asks a better question: from any point in the patch, how far is the nearest stop?

A 200 m grid over the box, and a distance from each grid point to the nearest access point. The answer is that 91.4% of the patch is within 400 m of a stop — roughly a five-minute walk — with a median distance of 149 m.

The worst-served point in the patch is 1069 m from anything, at 53.7752 N, 1.4921 W.

Degrees are not metres

Every distance on this page is in metres, and getting there took one function rather than one line.

Coordinates arrive in degrees. At this latitude one degree of latitude is about 111.1 km, while one degree of longitude is about 65.7 km, because the meridians have converged by 53.8° north. Pythagoras on raw degrees therefore stretches every east–west distance by a factor of about 1.69 and produces numbers that are wrong, plausible, and in no units at all.

atlaslib.distance_metres converts both axes before it measures. Hand-check 6, in chapter 2, tests it against a distance I worked out independently.

The figures

Every NaPTAN access point in the patch. Bus stops in blue, rail stations starred. The blank areas are the subject of the next figure.

Walking distance to the nearest access point, on a 200 m grid. Inside the blue line you are within 400 m of a stop; 91.4% of the patch is.

The same information as a curve. The steepness in the first 300 m is what a dense bus network looks like.

What it shows

• The patch contains 1,328 public transport access points — 1,314 bus stops and 4 rail stations — at 17.5 per km².
• 91.4% of the patch is within 400 m of a stop, with a median walk of 149 m. Access to something is close to universal here.
• The gaps are not distributed evenly: the worst-served point is 1069 m from any access point, which is a fifteen-minute walk before the journey starts.

Row counts

Every filter and every join in this chapter, with the row count on each side of it. A join that quietly dropped most of the patch would show up here as a percentage, not as an error.

Operation	Rows before	Rows after	Kept
Drop stops marked inactive	19,447	19,331	99.4%
Bounding-box filter to the patch	19,331	1,328	6.9%
Grid cells within 400 m of a stop	1,935	1,769	91.4%

Hand-checks

#	The claim	Checked against	Anchored outside the data	Result
1	Leeds City Bus Station is in the data	Its real address on Dyer Street (53.7975 N, 1.5372 W)	⚓ yes	pass
2	Leeds Rail Station is in the data	It exists, and it is the busiest station in the North outside Manchester	⚓ yes	pass
3	Every kept stop is inside the bounding box	The box coordinates themselves	○ no	pass
4	Stop density is credible for an English city	A plausible urban range of 5–40 stops per km²	⚓ yes	pass
5	Every stop appears in exactly one category on the map	The category counts, summed against the total	○ no	pass

Check 1. The nearest access point to the real bus station is Dyer Street (Leeds City Centre), 71 m away. Anything under about 150 m is the bus station itself or a stand inside it.

Check 2. Rail stations found in the patch: Burley Park Rail Station, Cottingley Rail Station, Headingley Rail Station, Leeds Rail Station.

Check 3. Tests the filter, not the data. It would pass just as happily on the wrong ATCO area.

Check 4. 17.5 stops per km² across 76 km².

Check 5. 1,314 bus + 4 rail + 10 other = 1,328, against 1,328 stops. The 'other' category is Rail station entrance (RSE), which the first version of this chapter drew nowhere.

What this chapter does not say

• A stop is not a service. NaPTAN records where you could board, not whether anything stops there, how often, or at what time of day. A stop served twice a day counts the same as one served every four minutes.
• Distance here is straight-line, not walked. Real walking distance is longer wherever a river, a railway or a dual carriageway is in the way — and this patch contains all three.
• The 400 m threshold is a convention, not a finding. It is roughly a five-minute walk on the flat, and Leeds is not flat.

Where the plan was wrong

Correction to the plan

Plan §2 said chapter 1 answers "where can you catch something". It planned a single query: ATCO area 450.

Area 450 contains bus stops only. Rail stations are in area 910. The plan would have produced a public transport atlas with no railway in it, and nothing in the build would have complained.

Fixed by fetching both areas. The hand-check that caught it was written before the code, in plan §5, which is the only reason it was ever run.

Correction to the plan

Plan §6 listed "a figure is wrong but the script succeeds" as a high risk. It happened on the first figure I drew.

The first version of the map had two categories, bus and rail. Ten stops in the patch are typed RSE, a rail station entrance, and belonged to neither. They were fetched, they passed the filters, they were counted in every row count on this page — and they were drawn nowhere.

Nothing failed. The count of stops was right. The legend added up to ten less than the total, which is the only visible trace it left, and I found it by reading the legend rather than by any check.

Fixed by making the categories a partition and adding check 5, which asserts that they sum to the total. The lesson is the one the risk table already predicted: open the figure and look at it.

Correction to the plan

The first coverage map was a broken figure that ran perfectly.

I set the colour scale to run from 0 to 2,000 m. The largest distance in the patch is 1069 m, so about nine tenths of the map came out in the same pale yellow and the figure separated nothing.

A colour scale is a claim about the range of your data. This one was a claim about a range that does not exist here. Fixed by fitting the levels to the data and using a scale with more contrast.

Correction to the plan

Prediction P1 said the patch would hold between 800 and 1,600 NaPTAN stops. The answer is 1,328.

Inside the range, but for the wrong reason: I was predicting bus stops without realising rail was a separate query. The prediction was right and the reasoning behind it was wrong, which is worth less than it looks.