The obvious way to build a property risk signal is to use everything you hold. More data, better model.
We deliberately do not.
Self-declared data is not weak, it is differently sourced
When a homeowner records that the boiler was serviced, that entry is genuinely useful to them and to a buyer. It is a claim, and it is labelled as one.
Feeding claims into a risk signal alongside verified work would produce a number that cannot be explained. Two properties with identical scores could be in completely different states, one documented by vetted contractors and one typed in by an optimistic owner.
So the signal uses the verified half only
Work that was completed through a vetted trader and recorded at the time. That is a smaller dataset and a defensible one, and it means a score can be traced back to specific jobs rather than to a general impression.
The properties with thin verified histories score as thin, rather than borrowing confidence from unverified entries.
Every read is logged
Every read of property-record data by a professional organisation is written to an access log. That is a control for the homeowner rather than a feature for you, and it is the reason this data is available to be looked at in the first place.
Early days, and we would rather talk about what is live now than describe a roadmap. If a signal you can audit is more useful to you than a signal that is merely large, that is the conversation worth having.