Criteria resolves lending questions against every lender at once.
Each query passes through a layered resolver. The broker asks in their own words. The system decides what kind of question it is, reads the published criteria of every supported lender, forms a position, and returns a broker-facing answer.
The query is classified: value, acceptability, policy, lender-specific, or composite. Product type inferred. Focus lender extracted if named.
For each lender, the system picks the sections of that lender's corpus most relevant to the query.
Selected sections are read in full. A committed position is formed: a number, a verdict, or a stance.
The position is converted into a broker-facing answer, grouped across the market, ready to read.
Ask in your own words.
A single criterion, a short question, or a full case. The system reads what you mean and resolves it against every lender at once. Name a lender and that lender is searched alone. Residential by default, unless the query implies buy-to-let.
Drawn from the criteria lenders publish to intermediaries.
Answers are drawn from the lending criteria that lenders publish and distribute to intermediaries, refreshed daily. Each lender's full criteria remain accessible from every result. The system surfaces the source, never obscures it.
- Ingestion cadence
- Daily at 06:00 UTC. All supported lenders refreshed in a single run; results committed only on full success per lender.
- Source material
- Lender-published criteria pages intended for intermediary use. Parsed structurally; content preserved verbatim where surfaced.
- Traceability
- Every answer carries a link to the underlying lender criteria section. Every response is reproducible on the data the system saw.























