Loading page
DG Score, confidence, data snapshot, data date, score version, model version, not financial, legal, or tax advice.
Every DG Score is assembled from named, dated, licensed UK government data. Every score keeps its receipts: which sources, which dates, which model version.
How a score is built
Official sources → five weighted risk categories → one DG Score
82
DG Score
The registry
Every source is listed with what it tells us and where it stops. The limits aren't fine print. They're the point.
What do homes actually sell for here, and how has that moved?
Every recorded sale in England and Wales, aggregated into local comparable prices and sales-volume trends.
Where it stops
Area-level comparables, shown as price bands and never passed off as the value of one property.
Official price growth, volatility, and drawdown history for the area, the backbone of Market Risk.
Where it stops
An index describes areas, not individual addresses.
Is the rent on this deal realistic, and what happens if it softens?
Local median rents, used to check an expected rent against what the area actually pays.
Where it stops
Local-authority averages appear as labelled suggestions, never written silently into your numbers.
The support-rent benchmark behind downside-rent stress tests: what income looks like if the market softens.
Where it stops
Stress context only. It adds 0% directly to the score.
Will this property cost money to make lettable, now or by 2030?
Energy band, floor area, and property details, matched to the address where a certificate exists.
Where it stops
Fallback matches are never treated as confirmed property-level facts.
The gap to EPC C, an estimated retrofit cost, and exposure to the October 2030 rental deadline.
Where it stops
No EPC match means lower confidence, never an invented retrofit bill.
What could the land itself do to this investment?
Flood zones and risk bands for England. Severe flood risk hard-caps the score, so there is no strong number on a floodplain.
Where it stops
A mapped-area signal, not a property-level flood certificate.
The equivalent flood-risk banding for properties in Wales.
Where it stops
A mapped-area signal, not a property-level flood certificate.
Whether the address sits in a mining reporting area, where survey, insurance, and lender friction may arise.
Where it stops
A reporting-area flag, not a ground-stability survey.
Distance to historic landfill and permitted waste sites.
Where it stops
Distance bands, not a contamination report.
Coastal erosion risk bands. Severe erosion risk will hard-cap the score, exactly like severe flood.
Where it stops
National mapping, not a site survey.
What is written into the title, and who is on the other side of it?
Lease dates and terms matched to the address, the foundation of Lease Risk scoring on leaseholds.
Where it stops
Only counts when match confidence is acceptable; a weak match stays out.
Companies House + land ownership data
In developmentLeaseplanned
Freeholder health checks: dissolved companies, liquidation, overdue accounts, and overseas owners behind the title.
Where it stops
An unmatched title reads as unknown, never as clean.
Article 4 directions, listed buildings, conservation areas, and green belt: the constraints that shape exits.
Where it stops
Flags and context only, zero score points: council-submitted data is too patchy to score honestly.
What is it like to own, let, and eventually sell here?
Street-level crime counts and category mix, a demand and desirability signal for the area.
Where it stops
An area-level proxy, not a property-level safety assessment.
The geography spine: postcode normalisation, coordinates, and the region joins every other source relies on.
Where it stops
Infrastructure only. It adds 0% directly to the score.
Live address autocomplete when you enter a property.
Where it stops
Address resolution only. It adds 0% directly to the score.
The official deprivation decile for the immediate neighbourhood, at a finer grain than any postcode average.
Where it stops
England only at first. Elsewhere it lowers confidence rather than guessing.
Sources marked 0% still matter: they carry provenance, stress context, and the geography joins that make every other signal trustworthy. Sources in development are design-approved and built in the open. Nothing here is promised vapourware.
The rules we score by
The same four rules govern every source above and every source we'll ever add.
When a source has no answer for a property, the score gets less confident. It is never quietly flattered. A gap in the data is reported as a gap, not scored as a pass.
Address-level data (an EPC, a matched lease) can prefill your deal, labelled with its source. Area averages appear only as clearly-marked bands and suggestions.
Everything here is free, licensed, attributable government data. No scraped listings, no portal feeds, no paid black boxes. If a source fails the licence check, it is dropped.
Each score records the data dates and model version it was built from. When you look back at a decision, you can see exactly what was known at the time.
In engineering now
The next model generation moves from good judgement to measured evidence.
Hand-picked thresholds are being replaced by percentile tables computed from ingested UK data, refreshed monthly and versioned so every historical score stays reproducible.
We score past data snapshots and compare the results with later area outcomes. This tests whether higher scores are associated with stronger outcomes without treating the backtest as a guarantee.
Instead of false precision, each score will carry a low–high range that narrows as your deal data completes. Missing data widens the range; it never inflates the number.
Every new source must pass a licence check. A signal only affects the DG Score after backtesting shows a stable, explainable relationship with later outcomes.