Two parcels sit 180 feet apart. A generic AI research tool reports them as neighbors. A zoning map reveals they are separated by a municipal boundary, a county road, and a state-maintained corridor with a different setback schedule. The "neighbor" parcel is, for any practical engineering purpose, in a different world.
This is the problem with radius-based proximity. It is the right answer to the wrong question.
What adjacency actually means
In GIS, adjacency is the relationship between polygons that share an edge or a vertex. Two parcels that share a boundary are adjacent. A parcel whose north line is the south right-of-way of a state highway is adjacent to that highway. A parcel 200 feet away, across that highway, is not adjacent to anything except its own street frontage.
For civil due diligence, adjacency is the load-bearing relationship. Setbacks reference frontage. Access rules reference the street the parcel touches. Buffer requirements reference the wetland at the property line, not the wetland down the block. The body of code that governs a site is built around what touches it.
Radius queries were the industry default because they are cheap. Pick a center point, drop a 300-foot buffer, pull whatever falls inside. The queries are fast, the answers feel complete, and the gap between "near" and "touching" disappears into the fog.
Where radius fails in practice
A developer we worked with in 2024 ran a 300-foot proximity query on a prospective acquisition in Cook County. The query surfaced nine nearby zoning districts. None of them were the one the site actually sat in. The parcel across the street had been rezoned in a spot action two weeks earlier. Three hundred feet was not enough. Three feet would have been.
Another case: a site in Palm Beach County, Florida, with frontage on a state highway. Florida DOT controls access management on designated state roads. A radius search showed a "local road" nearby, which the developer used to estimate traffic impact. The actual frontage was classified 29A, which forbids direct access for new commercial driveways. The project changed scope. The radius query never flagged the access class.
How Spatial Foundation reads the site
We classify every neighbor parcel and every adjacent road by the relationship it has with the subject. A parcel that shares a boundary is direct-adjacent. A parcel across a right-of-way is across-street. A parcel within a measured radius but without a shared edge is nearby. Each class carries its own implications for notification requirements, setbacks, buffers, and visual impact analysis.
Roads get the same treatment. The right-of-way is identified, the maintenance agency is identified, and where we have access to state DOT surveyor data, frontage linework is promoted from parcel-approximated to survey-grade. In Illinois, that means IDOT's route centerlines. In Florida, it means FDOT. The providers vary. The discipline does not.
Why this is foundational, not a feature
Spatial Foundation is not a search improvement. It is the substrate the rest of the system runs on. When the agent looks up setbacks, it asks the adjacency graph which frontage applies. When it checks if notification-to-adjacent-owners is triggered, it reads the direct-adjacent parcel list, not a radius. When it writes a site memorandum, it can say "the parcel directly north is zoned R-1, the parcel across State Route 83 is zoned C-2" without inventing a geometry it never verified.
Trust in a site report is downstream of trust in the geometry. We decided to pay that cost once, at the foundation, so every downstream answer inherits it.
What this means for your workflow
If you run pre-acquisition screening, the first check on any site is now the adjacency panel. Direct neighbors, across-street neighbors, and the full frontage roster render before any ordinance analysis runs. If the adjacency does not look right, nothing downstream will be right either.
For ongoing projects, the same layer feeds setback checks, utility easement analysis, and the neighbor-notification list used for public hearings. One source of truth for the geometry of the site. Everything else hangs off it.