Second city: Nairobi

The same data → screen → region_builder → method → eval → render pipeline as the Cape Town examples, run on a second country: Kenya kblock data clipped to the Nairobi metro bounding box, plus Open Buildings (data=nairobi_full), driven by the same composable BlockMetric variants used throughout this site. Each row below is its own run through that pipeline, reproducible end to end from the command recorded in its own meta.json.

variant	metric	blocks in region	OSM baseline
multiblock_depth_density	depth_density	7	yes

Shipped as-is, and why that is the honest framing

Nairobi's fabric is messier than Cape Town's, and these runs ship exactly as they came out rather than being tuned to look tidy. What does not carry over from the Cape Town examples:

• Region sizes. Nairobi's blocks are bimodal — some enormous, some tiny — so a building budget tuned against Cape Town's block-size distribution produces very different regions depending on which end of that distribution a run's seed block happens to fall on. No single budget fits every case.
• OSM footpath coverage. The depth and depth_density regions above carry a usable as-built osm_footpaths baseline alongside the synthesized methods; the density_compactness region has essentially no mapped footpaths, so its frontier grades only the synthesized methods, with nothing as-built to compare against — the table's OSM baseline column shows this directly.

The screens and metric behaviour carry over cleanly from Cape Town; region-growth tuning and OSM coverage are what differ, and the table above is where that difference actually shows up, rather than a claim made about it in this paragraph.

No ground truth here

The screen bake-off validates the screening stage against the City of Cape Town's own informal-structure survey — and that survey has no Nairobi counterpart. reblock.data.informal records the search behind that absence (the City's ArcGIS portal, openAFRICA, HDX, and OSM Overpass) coming up empty. Its precision/recall evidence is Cape Town only and does not extend to anything on this page: what appears above is what each method's frontier measures on its own terms, not a validated claim about which Nairobi blocks actually need reblocking.