docs/codebook.md

Building-use codebook

Status: Last update on 2026-10-09

Unit and question

One footprint polygon. Question: what is the main use of this structure, as it relates to electricity demand? Label what the evidence supports. When the evidence can't separate two classes, choose unknown.

Classes (keep in sync with LABEL_CLASSES in src/rtl/schemas.py)

class	definition	typical evidence	not this class
residential	Dwelling; people sleep here	Small-to-medium footprint in a cluster of similar buildings; no POI/tag; residential street pattern	A dwelling with a street-facing shop → mixed_shop_house
mixed_shop_house	Dwelling with a shop, kiosk, salon or small service in part of the building	Street-facing, on a main road or market street, a shop/POI matched to the footprint, elongated frontage	Stand-alone shop without a dwelling → commercial
commercial	Shops, markets, restaurants, bars, hotels, offices, banks, pharmacies, mobile-money and phone-charging kiosks	OSM shop=*/amenity=restaurant/bank…, Overture place, market cluster, large footprint on a main road	Workshops making or processing goods → productive_use
institutional	Schools, health facilities, government offices, community halls	OSM/registry facility match, large rectangular blocks in a compound, regular multi-block layout	Places of worship → religious
religious	Churches, mosques, temples, other places of worship	OSM amenity=place_of_worship / religion=*, Overture category, large single hall, often a cross-shaped or long rectangular plan	A school or clinic run by a church → institutional
productive_use	Small-scale production with motors or heat: grain mills, welding, carpentry, tailoring, agro-processing, cold storage, irrigation pump houses, water kiosks with pumps	OSM craft=*, industrial=*, Overture category, location near fields/irrigation or a trading centre	Retail without processing → commercial
industrial_warehouse	Factories, large warehouses, industrial-zone buildings	Very large footprint, industrial zone, landuse=industrial
ancillary	Structures with negligible electricity demand: latrines, detached kitchens, animal sheds, small stores	Very small (often < 10 m²) next to a larger building in the same plot	A tiny kiosk on a road → commercial
unknown	Evidence doesn't support a class	—	Never use as a "probably residential" bucket

Decision rules

1. Use a direct, matched tag or registry record when one exists, and note it as evidence.
2. Without a tag, use footprint size, shape, height, neighbourhood and road context.
3. Size alone never decides between commercial and institutional.
4. Under 10 m² with a larger building within 10 m → lean ancillary.
5. If two classes fit about equally, choose unknown and say which two.

Confidence

• high: a direct tag or registry match, or unambiguous context
• medium: several consistent indirect cues
• low: a single weak cue (labels with low confidence are kept but reported separately)

Decisions

• mixed_shop_house is separate from residential: shop-houses are common in Rwandan trading centres and add lighting, fridge and TV load.
• religious is separate from institutional, because its load profile is weekly rather than daily.