C•A•P•I•T•A•L

The Postcodes of the Inner London Postal District

Here's a map showing the 120 postcodes, with each code at the approximate centre of its postal district.

A map of the postcodes can be seen here [pdf].

 

The 'Capital' project is here

 

 

Let's try some abstraction. Here's those postcodes as a 12x10 grid. I used an AI (Claude) to do the heavy lifting, below, with the following prompt:

'can you take a map of the 120 inner london post codes and arrange them in as close an order as possible, west to east, north to south, as a 12 by 10 grid allows'

That's fun. It gives us a different perspective.

 

Let's get an AI to dig into this, may be it'll keep it away from that pesky 'paperclip problem'.

make a grid with no limitation on cells and rows showing the postcodes but cells can be empty. its still a schematic but one more in keeping with Beck

Our cheeful energy chomping chum has knocked out a 23 x 19 grid:

 

Now being an AI it has plenty to say about it's work. We should remain sceptical of the middle-management blah AI is happy to trot out but, it does show method:

That's the Beck logic applied properly: same projection as before (equirectangular, scaled by latitude for longitude compression), but instead of forcing every point into a filled 12×10 rectangle, I snapped each postcode to its nearest grid intersection on a much finer lattice and let empty cells stand. Where two postcodes would've landed on the same intersection, I nudged one outward to the nearest free point rather than distorting the whole layout.

Net effect: relative position and topology survive (north stays north, the west/east and north/south gradients are legible), but you can now see London's actual shape — the dense core clustering tightly around Zone 1, the outer boroughs spreading out unevenly (further south and west than north or east), rather than that shape being crushed into a uniform box. It's an SVG, so every point is just a labeled dot at fixed coordinates — straightforward to open in Illustrator/Figma and hand-adjust positions once you're pinning actual photos to each one.

 

You wouldn't want to be in that meeting... still, it may be useful as a hanging guide, the (x,y) co-ordinate is expressed as a pixel position within the CSS! Now I'd like to associate population numbers with those postcodes but let's start with a simple adjustment to the CSS to give a weighted value to each dot specifying one of five district sizes. And of course, when I say we, I mean, AI...

Postcode Area

Now let's have some of that all important AI backchat:

 

Using Voronoi tessellation on the same centroids I built for the map — each postcode "claims" the territory closest to its centroid than to any neighbour's, which is a solid proxy for actual district footprint when centroids are reasonably evenly distributed.

Caveat up front: the outermost districts (W7, SW20, SE2, E4, SE25 — the top 5) are inflated by an edge artifact — Voronoi cells at the boundary of any point-set balloon outward because there's no neighbour beyond them to cap the cell. Treat their absolute values as unreliable; everything from roughly rank 6 downward is a fair comparative read.

Largest (genuinely reliable, ranks ~6–15)
NW7 Mill Hill · N20 Whetstone · SE9 Eltham · SW14 Mortlake · E12 Manor Park · NW9 Kingsbury/Colindale · E18 South Woodford · N21 Winchmore Hill · SE12 Lee · SE20 Penge

Mid-large
SE28 Thamesmead · SW16 Streatham · N14 Southgate · NW10 Willesden · E6 East Ham · SE18 Woolwich · W13 West Ealing · SW19 Wimbledon · W5 Ealing · SW17 Tooting

Solidly mid-pack (bulk of the set, ~4–8 km² proxy)
E17, SE6, W4, SW15, N9, E16, SE3/SE7, SW18, NW4, W3, NW2, N3, N18, NW11, N15, E14, SE16, E11, N12 …

Small — dense inner core
Most SW1–SW10, most SE1/SE5/SE11/SE14/SE15/SE17/SE24, N1/N5/N7/N19, NW1/NW5/NW6/NW8, E1/E2/E3/E8/E9, W9/W11/W14, W8, WC1/WC2

Smallest
EC1, EC2, EC4 (the historic City core — tiny, tightly packed districts) and SW5 (Earls Court), the single smallest in the set.

The overall shape confirms the obvious intuition and gives it numbers: outer-ring districts (Mill Hill, Eltham, Whetstone, Mortlake) run roughly 30–40× the footprint of the tightest City-core districts (EC1/EC2/EC4), even setting the inflated edge cases aside. That gradient — huge quiet outer wedges shrinking to a dense tangle of tiny historic parishes at the centre — might itself be worth encoding visually (dot size, or stroke weight) on the schematic if you want the map to carry that information rather than just position.

Want the schematic regenerated with dot size scaled to this ranking?

 

Had to look up Voronoi tessellation here: wiki.

We're closing in on our population stats... next let's assocate dot size with population size. This took a while to get the information we need. The AI was unable to extract all of the data so I had to mess about with the postcode districts spreadsheet for a bit to get the figures. Take that AI! Here's the prompt

so now we have a solid set of data for postcode population. using that data id like to create a map like our 'area of postocde map' but using the differing size of dots to show population size.

 

Population Size by Postcode

Here's our map showing comparative population figures... innarestin'!

So now we've got all the data we need to calculate comparative density using our five circle sizes. Time for a new prompt:

now can we take the data from the postcode area map and the postcode population map and create a map showing population density?

As an afterthought I had the AI add concentic circles at 5km intervals moving out from Charing Cross. I also added that code to the two other maps above. And whilst I was at it we added a north/south and east/west axis centred on Charing Cross - just about!

Population Density by Postcode

Fabulous...

Here's the AI blah...

Density = population ÷ area (using the same capped area proxy as before, so the same five edge-artifact districts don't distort things), same positions, same five-tier quintile sizing — but this map tells a genuinely different story from the previous two, which is the point of building it.

Densest districts (small footprint, lots of people): N1 (Islington), SE1 (Southwark/Bermondsey), E1 (Whitechapel), N16 (Stoke Newington), W14 (West Kensington), N7 (Holloway), SE5 (Camberwell), SW9 (Stockwell). Classic inner-city terraced/mansion-block territory — small parcels, high occupancy.

Sparsest: EC3, EC4, E18, SW14, N20, NW7, SE2, SW20, W7, SE25 — a mix of the almost-unpopulated financial-district core (EC3/EC4) sitting at one end of the same list as the large, leafy outer-suburb districts (Mill Hill, Raynes Park, Whetstone) at the other. That's the interesting bit: density collapses two very different kinds of "empty" — office London and suburban London — onto the same small-dot end of the scale, which the population map alone or the area map alone couldn't show you.

 

Now let's take these density figures and rework our postcode 'periodic table'. We can include the names and take the quintiles and express them as shades of blue providing a heat map of density. Darker/denser, simps!

NW7 Mill HIll	N3 Finchley	N20 Whetstone	N12 New Southgate	N11 North Finchley	N14 Southgate	N13 Palmers Green	N21 Winchmore	N9 Lower Edmonton	N17 Tottenham	N18 Upper Edmonton	E4 Chingford
NW9 The Hyde	NW4 Hendon	NW11 Golders Green	N2 East Finchley	N6 Highgate	N10 Hornsey	N8 Muswell Hill	N22 Wood Green	N15 South Tottemham	E17 Tottemham	E11 Leytonstone	E18 Woodford & South Woodford
NW2 Cricklewood	NW3 Hampstead	NW5 Kentish Town	N19 Upper Holloway	N7 Holloway	N4 Finsbury Park	N5 Highbury	N16 Stoke Newington	E5 Clapton	E10 Leyton	E7 Forest Gate	E12 Manor Park
NW10 Willesden	NW6 Kilburn	NW8 ST John's Wood	NW1 NW Head District	N1 Northern Head District	E8 Hackney	E2 Bethnal Green	E9 Homerton	E3 Bow	E20 Olympic Park	E15 Stratford	E6 East Ham
W10 North Kensington	W11 Notting Hill	W9 Maida Vale	W2 Paddington	W1 West Head District	WC2 Covent Garden	WC1 Bloomsbury	EC1 Clerkenwell	EC4 The City of London	EC2 The City of London	E1 Eastern Head District	E13 Plaistow
W7 Hanwell	W13 West Ealing	W5 Ealing	W3 Acton	W12 Shepherds Bush	W8 Kensington	SW1 Knightsbridge	EC3 The City of London	SE16 Rotherhithe	E14 Popllar	E16 Victoria Docks & South Woodford	SE28 Thamesmead
W4 Chiswick	W6 Hammersmith	W14 West Kensington	SW5 Earl's Court	SW7 South Kensington	SW3 Chelsea	SE1 SE Head District	SE11 Kennington	SE17 Walworth	SE8 Eltham	SE7 Charlton	SE2 Abbey Wood
SW13 Barnes	SW6 Fulham	SW10 Brompton	SW11 Battersea	SW8 South Lambeth	SW9 Stockwell	SE5 Camberwell	SE15 Peckham	SE14 New Cross	SE10 Greenwich	SE3 Blackheath	SE18 Woolwich
SW14 Mortlake	SW15 Putney	SW18 Wandsworth	SW12 Balham	SW4 Clapham	SW2 Brixton	SE24 Herne Hill	SE21 Dulwich	SE22 East Dulwich	SE4 Brockley	SE13 Lewisham	SE9 Eltham
SW20 West Wimbledon	SW19 Wimbledon	SW17 Tooting	SW16 Streatham	SE27 West Norwood	SE19 Norwood	SE25 South Norwood	SE26 Sydenham	SE20 Anerley	SE23 Forest Hill	SE6 Catford	SE12 Lee

E20 - no data available

I ran this problem past another AI (chat GPT) and it made an interesting observation:

That is actually a fascinating optimisation problem: rather than asking "what does the postcode map look like?", we're asking:

What permutation of London's 120 postcode districts makes a regular rectangular graph behave as much as possible like London's actual spatial graph?

The grid above is therefore best thought of as a cartographic projection of London onto a rectangular graph, rather than a conventional map.

One particularly nice consequence is that the postcode prefixes cease to be meaningful categories. The grid discovers the geography underneath the postal bureaucracy.

Ok Computer, let's try:

can you merge the the three quintile circle diagrams to give us a readable population/area/density diagram for the entire map? maybe we need to use a different statistical device? perhaps we can show all three circles?

 

Well that's a little complicated. Let's have the AI tell us about it:

 

Here's the logic: each district gets two circles and a colour rather than three competing sizes:

• Grey ring (stroke only) = relative area — bigger ring, bigger district
• Filled dot inside it = population — bigger dot, more people
• Colour of that dot = density (your existing blue palette, light→dark, sparsest→densest)

This makes the interesting cases genuinely legible at a glance, in a way three separate maps couldn't show together: look for a big ring with a small, dark dot — that's a large area with relatively few people but packed tight where they are (patchy density within a big footprint), versus a small ring with a big, dark dot — tiny area, lots of people, obviously dense (your N1/SE1/E1 cluster). A big ring with a big, light dot is the opposite story — large and sparse throughout, no crowding anywhere (your outer-suburb Mill Hill/Raynes Park territory).

It's worth being honest that this is denser to read than any one of the three originals on its own — it rewards someone who spends a few seconds per dot rather than scanning the whole map at once. If that trade-off doesn't suit the piece, the alternative is a 2×2 or 1×3 small-multiples layout (three maps side by side, same positions, same size), which sacrifices the "everything in one glance" ambition but is much faster to read cell-by-cell. Happy to build that version instead if this one feels too busy once you're looking at all 120 together.

 

I wanted to check the LLMs working, so I ran the problem of where to put the 120 postcodes in our 12 x 10 grid against ChatGPT. At first it came up with the same solution as Claude. Tick, gold star. BUT... I mentioned the 'Seven Bridges of Königsberg' in my prompt and that sent our AI chum off in a completely different direction.

Something of the thought behind this project is rooted in my days working as a bicycle messenger. The ability to envisage a series of addresses and their connections in the most efficeint way to get between then lay at the heart of the job. The postcodes became a key to comprehending how one can move through the city as quickly and profitably as possible. Our AI pal thought about what we're trying to do and decided that what I would call 'nearness' is important. So:

the right optimisation target is not "make the grid look like a map" but "minimise the geographic cost of every horizontal and vertical edge in the 12×10 lattice."

Then it did this:

π min​ (i,j)∈E ∑​ d(p π(i)​ , p π(j)​ )

where:

• pkp_k = geographic centre of postcode district kk
• π\pi = permutation of the 120 districts into the 120 cells
• EE = the 218 horizontal/vertical adjacencies of a 12×10 grid
• dd = London-adjusted geographic distance.

 

Let's lay out the new grid and see what the AI has to say:

N14	N21	N13	N9	E4	E18	E20	E14	SE10	E13	E15	E11
N20	N12	N11	N22	N18	E17	E3	SE16	SE8	E16	E7	E12
NW7	N3	N10	N8	N17	E10	E9	E1	SE14	SE7	E6	SE28
NW9	NW11	N2	N15	N16	E5	E2	EC3	SE15	SE3	SE18	SE2
NW4	NW3	N6	N4	N5	E8	EC2	SE1	SE17	SE13	SE12	SE9
NW2	NW6	NW5	N19	N7	N1	EC1	SE11	SE5	SE4	SE6	SE20
W5	NW10	W9	NW8	NW1	WC1	EC4	SW1	SW9	SE22	SE23	SE25
W13	W3	W10	W11	W2	W1	WC2	SW3	SW8	SE24	SE26	SE19
W7	W4	W12	W8	SW7	SW10	SW15	SW18	SW11	SW4	SE21	SE27
SW14	SW13	W6	W14	SW5	SW6	SW20	SW19	SW17	SW12	SW2	SW16

We'll come back to that...

We need a river line to give our map a grounding in the familiar. Claude took twelve goes at getting the Thames onto our map. It's not perfect but it'll pass and it adds a sense of place.

And above the all important T-shirt design