← Back to methodology

Score validation benchmarks

We test the current scoring model (v2.3) against a mixed validation set of benchmark archetypes and a smaller number of measured/monitored homes. This page reports where the model aligns and where assumptions still dominate — it is not a compliance certificate.

Known limitations

  • Measured sample size is still small — most rows are benchmark archetypes.
  • Confidence is estimated from input completeness, not calibrated prediction error.
  • Architect design-brief quantities are separate indicative bands — not SAP/PHPP outputs.
ArchetypeEPC bandIndicative score bandSource
Victorian mid-terrace (uninsulated)D–E38–48benchmark
1980s cavity semi (partial loft fill)D45–55benchmark
Modern detached (Building Regs 2013+)B–C58–68benchmark
Deep retrofit (fabric + ASHP)B72–82benchmark
Monitored pilot home (anonymised)C52–58measured

Full automated validation report (MAE/RMSE) ships when the measured sample grows. Until then, use this benchmark table and the Passport validation lab for audience-specific projection tests.