How it works
Public registers, joined and measured, every week, for one council area — and a short list of things we will never do.
What we measure that you cannot ask for
Anyone can ask a chatbot "where are the development opportunities in Torbay" and get a plausible paragraph. Nothing on a site page is that kind of answer. Every site rests on at least one of five things a language model cannot do:
- Geometry — measurements on polygons and terrain: plot area, undeveloped area, the boundary that meets a road, median slope, what is within 50 or 100 metres.
- Joins — facts that only exist when registers are combined: a company owner and its Companies House status; an energy certificate and a plot and a street value; a parcel and every designation that touches it.
- Change — things that happened this week and only show if you watch continuously: decisions, lapsing permissions, appeals, price anomalies, company status changes.
- Outcomes — what actually happened to the sites we listed. Our data; nobody else's.
- Verification — a surveyor has looked at it.
Every site page says which of these it rests on. A site that rests on none does not go in.
The sixteen signals
| Family | Examples | Rests on |
|---|---|---|
| Geometry | backland or infill capacity behind an existing building; bungalow on a large plot in a high-value street; corner plot with side access; large pre-1929 villa with conversion precedent nearby; garden-land assembly across several parcels; vacant urban land with no structures on LiDAR | measurement |
| Joins | company owner in liquidation, administration or strike-off; tired commercial stock within the Class MA floor-area limit; long-held property in poor condition on a large plot; public-sector-owned land with capacity; tourism stock outside the tourism-retention areas, in poor condition or recently sold | register × register |
| Change | permission granted 30–36 months ago with no evidence of commencement; appeal allowed within 500 m in the last year; sale in the last 90 days at 70% or less of the sector's £/m²; draft Local Plan allocation adjacency (context only, never a trigger) | watching the registers |
Each signal has a threshold you can read, a unit test on synthetic parcels, and a list of the ways it can be wrong. Signals are withdrawn on a page when a check fails — an institutional freeholder, a leasehold block, a highway or amenity parcel, a rural postcode too coarse to locate an appeal — and the page says which were withdrawn and why.
Scoring
score = value × feasibility × timing × confidence, each 0–1. Value from plot capacity and street context, or from the permitted scheme, or from converted-stock comparables within 500 m. Feasibility from slope bands, heritage and tree constraints, access and flood risk — a fatal flag (flood zone 3 with no permission, a listed building) leads the page or excludes the site. Timing is highest for dated triggers. Confidence reflects what data the parcel has, and is capped while any driving fact still carries a "verify" note. Every score comes with a one-paragraph explanation built only from the evidence fields.
Data sources
| Dataset | Publisher | Used for |
|---|---|---|
| INSPIRE Index Polygons | HM Land Registry | parcels (indicative extents of registered titles, not precise boundaries) |
| Price Paid Data | HM Land Registry | sales, street and sector £/m², timing, new-build ledger |
| UK Companies that own property; Overseas companies that own property | HM Land Registry | company ownership — company names only, never individuals |
| Company profiles | Companies House | company status (liquidation, administration, strike-off, overdue accounts) |
| Energy Performance of Buildings Data (domestic and non-domestic) | MHCLG | floor area, type, age band, rating; commencement evidence |
| Planning applications and decisions | councils, aggregated by PlanIt | full register since 2002; lapsing permissions; precedent |
| Appeal decisions | Planning Inspectorate | allowed appeals nearby |
| Conservation areas, listed buildings, TPO zones, Article 4, flood zones, designations | planning.data.gov.uk (with Historic England and Environment Agency data) | constraint flags |
| Open UPRN, OpenMap Local, Open Roads, Boundary-Line, Open Greenspace, Built Up Areas, Open Names | Ordnance Survey | addresses, buildings, roads, the territory boundary, exclusions, maps |
| National LiDAR Programme, 1 m DTM/DSM | Environment Agency | slope, aspect, structures, frontage obstructions, vegetation structure |
| Local Plan, Neighbourhood Plans, Housing Land Supply, Authority Monitoring Report | Torbay Council | policy context, tourism-retention areas, the planning climate |
| Greater horseshoe bat consultation zones | Devon County Council | ecology cost flag |
Licences and the exact statements each publisher requires are on the attributions page. Coverage differs by council — a territory page states which layers its council has and has not published.
Territories and exclusivity
A territory is one local planning authority. Each carries three Founder seats, then up to five Standard seats — or one Exclusive seat instead. Every seat in a territory receives the same brief on Monday, with one difference: when a seat marks a site pursuing, that site is withheld from the other seats' briefs for thirty days, and the other seats carry a different site in its place. Pursuing is a reply with the site's reference. It costs nothing and it tells us — the only way anyone can — that a real developer judged the site worth chasing.
New territories open by demand, in the order of the list, and take about a week to stand up from the national datasets plus the council's own layers.
What we never do
- Name a private individual. Owners are shown as a company name and number from the public register, or as "private individual(s)". No table of individuals exists.
- Contact an owner, or help you to. No letters, no lists, no "owner outreach". The Land Registry licences forbid using ownership data to approach proprietors, and the product is the site, not the mailshot.
- Imply a site is for sale. A site in the brief is a site worth a call; nothing more is claimed.
- Scrape portals. No Rightmove, Zoopla or OnTheMarket data; council planning documents are read politely and cached.
- Let a language model assert a fact. Ownership, planning history and constraints come from data; the model writes, it does not know, and every number in a write-up must exist in the record or the page is rejected.
- Pad a quiet week.