Collateral valuation data for lending decisions.
Evaluate value, confidence and range for origination, second-value review and portfolio monitoring. A completed test record must state the cohort, date, baseline, coverage and performance by confidence band.
Pretira supplies an estimate, confidence output and range. A lender-run evaluation then decides whether any confidence threshold is stable enough for routing and which cases still need people.
What a publishable result must contain.
A headline accuracy number is not a certificate. The rows below define the record Pretira should publish once a completed cohort and result are available. They do not describe a result that already exists.
This is a publication specification, not performance evidence. A certificate becomes current only when a completed cohort, results, model version and material limitations are published together.
Sold prices, not opinions.
The model estimates sale value. A defensible lender evaluation freezes predictions and compares them with a dated cohort of completed sales, with the metric and limitations recorded. The first public Pretira test certificate is still pending.
Confidence as a routing input.
Confidence can be evaluated as a routing input, but it is not a probability that an individual estimate is correct. Set high, medium and low thresholds only after testing them on the intended lender cohort.
The second-value pattern.
A second valuation can expose cases where one model has strong evidence and another does not. Pretira gives lenders another governed signal for routing, with the overlap measured on the lender cohort rather than promised in general.
Test high-value properties separately.
Premium stock should be reported as its own cohort rather than buried inside a national average. A completed testing pack must show its dates, sample size, baseline, coverage and error distribution before a lender relies on it.
Rental view included.
Rental value estimates and days-to-let context for BTL propositions, from the same call. One request returns the sale view and the rental view together.
Confidence and range can help identify records for review, but their labels do not prove accuracy. Test the thresholds on the intended cohort, then route exceptions according to lender policy.
The four questions we always get.
Valuation outputs are decision-support estimates, not formal valuations, and no substitute for professional advice or survey where required.
Test the route, then publish the evidence.
Start with the synthetic sample, a controlled sample or a time-boxed API trial. A lender-run completions test can stay inside your controlled environment, with cohort, baseline and results recorded together.