Spatial language: scope, uncertainty and map resolution

Last updated: 7 Oct 20265 min read

tutorial

Keep qualitative location phrases as uncertain regions and distinguish a resolved anchor from a verified target location.

Carry the relation through resolution

A map pin is a point estimate. “South of the rail yard” is a directional region, and “between the yard and the clinic” depends on two anchors and an unstated corridor width. Preserve the original phrase and its relation frame even after selecting anchor IDs. If the product needs a point for display, mark it as an approximation and avoid using it for eligibility or emergency dispatch. Toponym frames keep the anchor distinct from the target.

Separate precision from certainty

A gazetteer can provide coordinates with many decimal places for an anchor while the described event remains vaguely located. Store anchor resolution confidence, relation interpretation confidence and geographic extent separately. A location may be certain at district level but unresolved at street level. Expose this distinction in downstream filtering and maps. Do not convert uncertainty into false precision through rounding or a default centroid.

Choose geometry only under policy

An explicit “within 3 kilometres” can become a distance test once coordinate system, Earth-distance method and measurement convention are defined. “Near” requires a task-specific policy or a clarification. Polygon containment requires a boundary version and a choice about points on the edge. Beware of degrees-versus-metres errors and antimeridian crossings. The small code sample below checks whether a numeric-distance query is admissible; it intentionally leaves geodesic computation to a spatial service.

Audit geographic errors

Create fixtures for duplicate place names, outdated facilities, nested regions, border points, ambiguous directions and missing anchors. Score whether the system asks for clarification when appropriate, not just distance error after it guesses a point. Review the errors by region and document style. A correct anchor with the wrong relation can send a responder to the wrong side of a facility. The field report project tests the release boundary.

Implementation

def admit_spatial_filter(frame, allowed_anchor_ids, maximum_radius_km):
    if frame.get("state") != "bounded":
        return {"state": "review", "reason": "uncertain-extent"}
    if frame.get("anchor_id") not in allowed_anchor_ids:
        return {"state": "deny", "reason": "anchor-scope"}
    radius = frame.get("radius_km")
    if isinstance(radius, bool) or not isinstance(radius, (int, float)) or             not 0 < radius <= maximum_radius_km:
        return {"state": "review", "reason": "radius-policy"}
    return {"state": "eligible", "anchor_id": frame["anchor_id"],
            "radius_km": radius}

bounded = {"state": "bounded", "anchor_id": "depot-47", "radius_km": 3}
assert admit_spatial_filter(bounded, {"depot-47"}, 7)["state"] == "eligible"
assert admit_spatial_filter({**bounded, "radius_km": True},
                            {"depot-47"}, 7)["state"] == "review"

Performance and operating cost

With a hash-backed anchor set this gate takes expected O(1) time and space. Computing geodesic distances or intersecting polygons is a separate operation whose cost depends on the spatial index and candidate count. Eligibility here does not certify that a target lies inside a region.

Common Mistakes

Displaying a vague relation as a precise event pin.
Confusing anchor-coordinate precision with event-location certainty.
Comparing geographic degrees to a radius in kilometres.
Ignoring boundary versions when checking containment.