The process
How a dataset gets built.
Read the licence in full
Before any code is written, we read the terms attached to the register — not the summary page, the actual licence. Public availability is not permission. Some registers are open for commercial re-use; some carry a copyright notice with an enumerated list of permitted purposes; some prohibit exactly the use that would be most profitable.
Where the position is unclear, we write to the rights-holder and ask, before building. A written answer costs nothing and takes weeks. Discovering the answer after you have customers costs refunds and credibility.
Take the whole register, not a sample
We ingest the complete published dataset, not a scrape of a search interface. That matters for coverage — a scraped product silently misses whatever the search box would not show you, and neither you nor we would ever find out.
Reduce to one current record per property
Registers accumulate. A single property may carry several certificates lodged over fifteen years, under addresses written four different ways. We normalise the addresses, resolve them to a single property, and keep only the latest record — so you are never sold the same address twice, or contacted about a rating that was superseded a decade ago.
Filter to the population that concerns you
Then the trade-specific filter runs: tenure, rating band, expiry window, fabric condition, fuel type. This is where a general property database becomes a working list. It is also where most of the judgement sits, and where we would rather return 400 properties that are genuinely relevant than 4,000 that are technically matching.
Score it, so the list arrives sorted
Every property gets a priority score built from the things that actually predict imminent spend: how poor the current rating is, how close the compliance deadline is, how large the achievable improvement is, and what the certificate says the owner would save each year.
Publish with the notices attached
Every file — paid or free sample — is generated by a single process that refuses to produce output if any required notice is missing. The copyright and attribution notices, the data vintage, the accuracy caveats and the end-user terms travel with the data, every time.
That is a licence condition rather than a courtesy, and it is enforced in code rather than by memory.
The refresh cycle
Monthly, on the first, whether or not anything interesting happened.
Rebuilt, not patched
Each month the product is rebuilt from the current published release rather than incrementally edited. Patched datasets drift; rebuilt ones cannot.
Withdrawals are honoured
When a certificate is removed from the source register, it is removed from sale — not merely dropped from the next file. It goes on a permanent suppression list.
Every file is dated
Each delivery states the exact period of the register it was compiled from, so you always know how old the oldest record could be.
Where we stop
Things we will not build, at any price.
No owner identification
No owner names, no correspondence addresses, no company-to-property matching, no skip-tracing, no "find the landlord" upsell. Ownership data sits under licences that prohibit marketing use, and there is no version of that feature we could build and still describe this product honestly.
No contact-detail enrichment
We do not append phone numbers or email addresses to properties, and we do not partner with anyone who does. If you need to reach an owner, that is your research on your own lawful basis — and our terms say so explicitly rather than leaving it vague.
No shared or resold enquiries
We are not a lead-generation agency. Nothing we sell is an enquiry, a referral, or a name passed to five installers at once. It is a compiled dataset, priced as one.
No claims the data will not carry
We will not describe a property as non-compliant, or its owner as breaking the law, and our terms prohibit customers from doing so either. The register supports "the latest certificate says X" and nothing stronger.