Questions

The engine is a deterministic constraint model: explicit rules, published rates and arithmetic, wired together so one change re-derives everything. That is deliberate — it is what makes the outputs testable, explainable and insurable. We use language models only to help you describe a project and to explain results in plain English; they never compute a number. Machine learning will calibrate the model once there is outcome data from completed projects.

Concept-stage ranges, shown as P20–P80 with the uncertainty class of each package. The prototype uses seed rates from published guides; the product will use a licensed cost dataset and calibration against real completed projects. Sketchr is not a quantity surveyor's estimate and never shows an unqualified single figure.

No. It is a rule-based reading of the householder permitted-development classes and your site's designations, with the rule cited for every flag. It tells you whether an application is likely to be needed and how long it typically takes; it does not predict approval and it is not a substitute for a planning consultant or a lawful development certificate.

No. You bring a brief or a design (yours or your architect's) as parameters; Sketchr tells you what it does to your money, timeline, planning position, running cost and carbon, and what to change if something breaks.

The energy figure is an indicative fabric estimate, not an EPC; an EPC is produced by an accredited assessor. Structural flags tell you when a beam or an engineer's calculation is needed; they are not calculations.

In the prototype on this site, nowhere — it runs entirely in your browser. In the product, data is held in the UK, encrypted, used only to run your project and, with your consent, anonymised for calibration.

That is the roadmap. Sketchr produces structured phase-and-cash-flow analytics; it does not make lending decisions. We are looking for one lending partner for the pilot.