Methodology · the formula, the failure mode, the limit
Every figure, and exactly how it was produced
This page shows the exact method behind every measurement — the formula, the failure mode, and what each number cannot tell you. Most location tools publish the number and keep the method. Nothing here is a score, and nothing here is a recommendation.
Data sources: WorldPop (population, 3 countries only), Overture (places), OpenStreetMap (POIs, roads), ESA WorldCover (land cover). Coverage disclosures are noted in each section where applicable.
Where the method stops
LocateRight produces raw observations and transparent derived measurements: counts, distances, densities, coverage. Each one is traceable to a source, a version and a formula. The platform then stops.
There is no scoring stage, no ranking stage and no recommendation stage, and this is a deliberate design constraint rather than a missing feature. A single number that says where to open would have to embed assumptions about your business model — price point, daypart, catchment behaviour, competition response — and those assumptions differ between a coffee bar and a pharmacy. A composite score would hide them inside a figure that looks decisive.
So the judgement stays where it belongs: with you, holding evidence that can be argued with. A useful consequence of this constraint is that a measurement can tell you when a location is not viable, which a score tuned for a median business often will not.
Composite scores, ratings, rankings or suitability grades produced by the platform.
Fields attached to every measurement: source, dataset version, retrieval timestamp.
Distinct causes of an empty result, reported separately instead of as one blank map.
Each measurement, as implemented
What follows is the method the platform actually runs, not an aspirational description of it. Where a figure is a sampled estimate rather than a census, that is stated in the same table rather than in a footnote.
| Figure | How it is computed | What it is not |
|---|---|---|
| Population | Sum of stored WorldPop cells of roughly 100 m resolution falling inside the analysis radius. The radius is reported with the value. | A headcount, a customer base, or a forecast. It is a modelled grid estimate — strong at city scale, weak at street scale, and silent on who the people are. |
| Place and business density | Records from Overture Places and OpenStreetMap POI, de-duplicated by record so the same place is not double-counted across sources, counted within the radius. Density is that count divided by the circle's area: π × (radius in km)². | A measure of competition or of market size. It also measures how thoroughly a region has been mapped, which is not the same thing as how many businesses exist. |
| Land cover | Classes counted inside the radius from the stored ESA WorldCover 10 m tile covering the point, reported as a percentage share per class with the dominant class named. | Land use. It does not know what a building is used for, whether a plot is let, or whether a site could trade. |
| Building footprints | Count of footprint records inside the radius across the Microsoft and Google Open Buildings stores, with the dataset version recorded alongside — the Google Open Buildings reference sets out what a footprint record is. | A building census. Footprints exist only where a publisher released them, so a zero is a coverage gap, not a finding. |
| Drive-time accessibility | A sample of origin points in the study area is routed to the site through a routes matrix. Reported as average, fastest and longest travel time, the share of sampled points reachable within set minute bands, and a distribution. Unroutable points are counted and excluded; statistics cover the routed share only. | An isochrone. The result is an explicitly labelled sample, not a smooth catchment polygon, and it says nothing about who travels those routes or when. |
| Competitor listings | Category-matched place records within the radius with straight-line distance from the point. Co-location and missing-category checks compare which categories are present in the area against an expected set. | A ranking. A close competitor list is a list of nearby records, not a judgement about which of them threatens you. |
| Opening hours | Published hours where the source record carries them, shown as published. | Verified or current hours. The value is a claim by the source and is absent for most records. |
Provenance: the three fields on every figure
Every measurement and every exported record carries the same three fields. They are the mechanism that makes the rest of this page checkable rather than merely asserted.
Because the version is recorded rather than assumed, two analyses of the same point on different dates can be compared honestly: if a figure moved, the version stamp distinguishes a change in the world from a change in the upstream dataset. Without that field the two are indistinguishable, and a published change cannot be explained — only observed.
The same fields are attached to exports, so a figure keeps its origin after it leaves the platform. Full source and licence detail is on the data sources page; if a figure looks wrong for a location you know, report it on the contact page and the provenance fields above are what make the report checkable.
How the method fails, and how that is surfaced
A method that cannot fail is not a method. Each of these is a real failure mode in the pipeline, and each is reported rather than hidden behind a default value.
No stored raster covers the point
Population, land cover and building blocks each return an explicit
NO_DATA_FOR_LOCATION state with the reason attached, instead of a zero. A
zero would be indistinguishable from a measured absence, which is a materially different
claim.
The region is not ingested yet
Data may exist upstream and simply not be loaded here. This is stated as its own message rather than folded into "no results", because the remedy is completely different: one is a gap in our pipeline, the other is a gap in the source.
The category is not in the loaded layers
A search can return a valid area with no records of the requested type. That is a filter result, not an absence of the category, and the interface says which case applies.
Points cannot be routed
In the drive-time probe, unroutable sample points are counted and excluded rather than dropped silently, and the statistics are reported over the routed share only. A drive-time figure that hid its own unreachable origins would overstate coverage.
Upstream data is uneven
Completeness tracks mapping effort, not reality. Place density inherits this directly, which is why it is reported as a count with its radius and source attached rather than as a benchmark against other areas.
No gaps are ever filled
The platform does not interpolate, smooth or synthesise missing values. An empty result is information about the data's coverage, and replacing it with an estimate would be the one genuinely dishonest thing a mapping product could do.
Reading the map honestly
The map draws distance rings around a point. They are scale references, labelled in metres and kilometres, and they are not travel time. A circle on a map is a convenient fiction that roads routinely contradict — which is exactly why the drive-time probe exists and why it reports a sampled distribution instead of drawing a smooth blob that looks authoritative.
When the two disagree, believe the drive time. The ring is a radius; the routes matrix measures what a journey actually costs. Treating them as equivalent is the most common way a catchment gets overestimated, and it usually happens silently.
What no method here can answer
Some questions are not reachable from this data, and saying so is more useful than a plausible-looking proxy.
- Will this site trade? Revenue depends on your price point, daypart, capacity and brand. No geospatial dataset contains any of those.
- Who are these people? Population grids carry counts, not identity, income or intent.
- How much footfall passes this door? That is mobile-device-derived measurement, and it is not something this platform collects or estimates.
- Which competitor will react, and how? A nearby listing is a record, not a strategy.
- What will this area look like in three years? The method measures a release of a dataset, and is explicit about which release.
A tool that answered these with a confident number would be answering a different question from the one you asked, and you would have no way to tell from the output.
Methodology: FAQ
Does LocateRight produce a score or a recommendation?
No, and this is a design constraint rather than a missing feature. The platform produces raw observations and transparent derived measurements — counts, distances, densities, coverage — and stops there. It does not rate a neighbourhood, rank streets, forecast sales, or recommend a location. A single number that tells you where to open would hide exactly the assumptions you need to argue with, and those assumptions differ by business model in ways no universal formula can settle.
How is the population figure calculated?
Population is the sum of stored WorldPop cells falling inside the analysis radius, at roughly 100 metre cell resolution, and the radius used is stated with the result. It is a modelled gridded estimate derived from census and survey inputs, not a headcount. It is reliable at national and city scale and weak at street scale, and it says nothing about who those people are or whether they are your customers.
How is place density calculated?
Place records are gathered from the Overture Places and OpenStreetMap POI sources, de-duplicated by record so the same place is not counted twice across sources, and counted within the radius. Density is that count divided by the area of the circle, which is pi times the radius in kilometres squared. The result is places per square kilometre at the stated radius, and it inherits the completeness limits of the underlying open data: a region that has been mapped more thoroughly returns more records regardless of how many businesses actually operate there.
How is land cover measured?
Land cover is ESA WorldCover at 10 metre resolution. The stored tile covering the point is identified, classes are counted inside the radius, and each class is reported as a percentage share of the total, with the dominant class named. It is a surface-cover classification, not land use: it does not know what a building is used for, whether a plot is let, or whether a site is commercially viable.
How are building footprints counted, and why can the count be zero?
Footprint records inside the radius are counted across the Microsoft and Google Open Buildings stores, and the dataset version is recorded with the result. A count of zero is reported as a coverage gap rather than as a measurement: footprints only exist where a publisher has released them, so a region with no published footprints returns no data legitimately. A zero here is a statement about the upstream open data, not about the buildings that are physically present.
How does the drive-time measurement work, and what are its limits?
Rather than drawing smooth isochrone polygons, LocateRight routes a sample of origin points to the site through a routes matrix and reports what came back: average, fastest and longest travel time, the share of sampled points reachable within set minute bands, and a distribution. Points that cannot be routed are counted and excluded, and the statistics are reported over the routed share only. The result is explicitly labelled as a sample rather than a full population, which makes it checkable instead of merely plausible.
What does an empty result mean?
An empty result is information, and LocateRight distinguishes three causes rather than showing a blank map: the region is not yet ingested, the source has no coverage at that point, or the requested category is not among the loaded layers. The platform never fills a gap with synthetic or interpolated data. If a figure is missing, the honest answer is that the data is missing, and the interface says which of the three cases applies.
How do I verify a figure I have seen?
Every measurement carries three fields: the source it came from, the dataset version it was computed from, and the timestamp it was retrieved. That provenance is attached to exports as well as displayed on screen. Because the version is recorded, two analyses of the same point taken on different dates can be compared honestly — if a number moved, the version stamp tells you whether the world changed or the upstream dataset did.
Now check the method against a real place
You have the formulas and the failure modes. Put a candidate location on the map and verify them yourself: every figure arrives with its source, its dataset version and its retrieval timestamp attached.
Working with an AI agent that asks for a machine-readable index? /llms.txt lists every public page on this site with a one-line summary of what each one covers. It is a convenience file, not a Google ranking factor.