Summary
A property-analysis platform that matches HM Land Registry transactions with Energy Performance Certificates, scores every matched property from its energy-efficiency features, and ranks the results so analysts can find and assess Stamp Duty cases.
Context
Stamp Duty cases turn on property facts: what a property sold for, when, and what condition it was in at the time. In England and Wales two public datasets hold most of those facts: HM Land Registry price-paid records and the domestic Energy Performance Certificate register.
The client runs a litigation process in which those facts have to be assembled and assessed case after case. Doing that by hand does not scale, and it makes each assessment depend on who prepared it.
What N-Thesis delivered
N-Thesis built the platform behind that process: the data loading, the matching, the scoring and the portal analysts work in.
- Bulk loading of EPC data from the public EPC API and of Land Registry price-paid transactions
- Matching of each sale to the energy certificate that was valid before it
- Feature-level scoring of walls, roof, windows, floor, heating, hot water and lighting, combined into property scores
- A portal to search and filter matched properties by date, postcode, district and price, with the features behind each score
- Score tuning: analysts adjust feature scores individually or in bulk, and every property is re-scored automatically
- Deployment to a cloud server, provisioned and configured as code
- Public datasets
Land Registry price-paid transactions and domestic EPC records, loaded in bulk.
- Address matching
Each sale matched to the certificate valid before it, using normalised addresses and postcodes.
- Feature scoring
Energy-efficiency ratings of each building feature, weighted and combined into property scores.
- Case analysis
Ranked, filterable properties for the client’s Stamp Duty case work.
The difficult problem
The two datasets describe the same properties without sharing a reliable key. Addresses are written differently in each, so matching depends on normalising addresses carefully, joining on postcode, and checking that the certificate predates the sale. Records that do not match are kept rather than silently dropped.
Volume was the second difficulty. The EPC API limits how many results a single query can return, so the loader splits the work by district and by month, fetches pages in parallel, and records which ranges are complete so that a restart never fetches the same data twice.
The third was keeping the scoring explainable. Scores are built from individual building features with visible weights, so an analyst can see why a property ranks where it does, and change the weighting when the case work shows it should change.
Results and delivered capability
Analysts work from a single ranked view of matched transactions instead of assembling property facts by hand, and the scoring can be tuned as the case work develops.
We publish outcome figures for client work only once the measurement behind them has been documented and approved by the client. This page describes the delivered capability.