Parcel proposals from open footprints: boundary generation for informal settlements, and what deployment still needs

Most of the world's people hold land without a registered boundary, and in many places that record is the gate to essential services: streets, water, power and sanitation. A boundary can be surveyed on the ground, detected in imagery, or generated from what has been built. In the Netherlands, with complete cadastre and good imagery, overlaying cadastre on the topographic map shows that 27.8 percent of boundary length coincides with no physical feature that detection could find; where cadastre and imagery are scarce the share is plausibly higher, and in informal settlements most of all. We report a model that generates parcel boundaries from building footprints: a learned boundary field, decoded into parcels with or without a geometric prior, and measured against a published morphological tessellation as the floor it must beat. In all sixteen geographies a model never saw it places boundaries better than that floor, while its effect on how well each parcel's area is matched is small, and in places negative. On Mayotte, a French department off East Africa whose open cadastre covers informal fabric, the correction transfers zero-shot from an entirely Northern corpus, and it works only partly where it matters most: in the densest settlement 55 percent of the parcels cannot be reached from footprints at all. With footprints anyone can download, between three fifths and all of the correction survives across three informal scopes on two continents. A proposal costs under six thousandths of a cent in compute, from open data, against about six to fifty dollars per parcel for registration programmes; a proposal is not a registered boundary, and the figures show only that compute is no longer the cost. Held to the seven fit-for-purpose principles, generation is affordable and attainable, unreliable if mistaken for cadastre, and does nothing by itself for inclusion and participation; today's web technology lets a whole community check and correct a proposal from any phone, where before only experts could, and that correction loop, which gives the people who hold the land the final say, is what deployment still needs.