Bentuk Kota

The problem

People who study city street patterns mostly use one measure: orientation entropy, from Geoff Boeing's 2019 paper. It is computed on the driving network. In Indonesia that misses most of how people actually move. A kampung is threaded with gang, narrow pedestrian alleys that never appear on a car map. A gated perumahan is easy to drive into, but it has one or two entrances and dead ends everywhere. Measured by car alone, both come out wrong, in opposite directions.

I built Bentuk Kota for urbanists, planners, OpenStreetMap mappers and curious readers. It measures the same disc of the same neighbourhood twice, once as a driving network and once as a walking network, and puts them side by side. The gap between the two is the finding.

Kampung versus perumahan carries class overtones in Indonesia. So I set one rule for the whole product: it describes and never scores. There is no ranking, walkability grade or liveability index anywhere.

What I built

I wrote the street maths myself

The measurements are the project, so I didn't import a graph, chart or map library. Building the networks, reading street bearings, and computing entropy, detour ratios and junction counts are all my own code. I wrote it from the paper, with its sections cited. The measuring code takes numbers in and gives numbers out, so the same code runs in the data pipeline and in the tests.

Tests use streets with known answers

Nobody knows the true entropy of Bendungan Hilir, so the tests don't pretend to. They build networks whose answers are known in advance: a perfect grid, the same grid rotated 29°, a random network and a pure tree. Then they check that the measured numbers come back right. Every chart, synthetic or real, must also pass basic maths checks, such as looking the same when turned 180°.

The biggest assumption is a switch readers can flip

Which OpenStreetMap tags count as drivable or walkable changes every number on the site. I put that choice in one documented place and exposed it on an assumptions page, as a switch between three mappings. Readers can see how much each conclusion depends on it.

I picked sites by measuring map coverage

My first plan was a hand-picked list. Two rounds of guessing produced 10 usable sites from 49 guesses, and the misses proved nothing. So I wrote a sweep across 56 cities that ranks candidate areas by how many footpaths are actually mapped. It found 14 of the final 33 sites, including the first ones in Papua, Lombok and Kalimantan.

I wrote rules for where coverage misleads

Mapped footpaths can't tell an alley from a garden path. Bogor's best-covered candidate, at 40.9%, was a botanical garden. Jakarta's best was 78% inside one park. I turned those lessons into rules. I also added a check that the survey must predict the pipeline's result within one percentage point, which caught two sites entered with stale coordinates.

Result

Bentuk Kota is live and public, in Indonesian and English. It covers 33 sites, each measured as two networks under three tag mappings, chosen from 126 candidate centres surveyed across 56 cities; 22 of those candidates were rejected and published anyway, with their reason stated rather than deleted. Every site shows how well its alleys are mapped, and 9 of the 33 are flagged for thin coverage so missing map data can't fake the finding.

That limit reaches the headline comparison. Every gated-housing candidate came back thin, at 0.9–4.4% coverage, so the kampung versus perumahan comparison rests on two readable gated sites out of three. The page states that number. The site makes no network calls once it loads. I built the first version in eight days in August 2026, then kept working on it into late September.