The configurator is the FEP engine with the lid off. It lets a commissioner move all twenty-one signal weights and watch the district ranking recompute in real time — a deliberate act of methodological transparency.
Most risk-stratification products hide their weighting behind a proprietary wall and ask to be trusted. We take the opposite view: in a public-sector context, a score that cannot be interrogated should not move public money. Exposing the weights converts the model from an oracle into an argument — one a commissioner can win, lose, or amend.
It also surfaces the single most important truth about any composite index: the ranking is a function of the weights. Two reasonable people with different clinical priors will produce different maps from identical data, and they should be able to see exactly where and why they diverge.
Each signal is pre-normalised 0–100 within signal. The configurator computes a weighted sum, renormalises across zones and re-ranks. An uncertainty flag is attached to any zone whose ordering would flip under a small perturbation of the weights — a direct, visible expression of how robust each placement is.
The built-in Ada explainer narrates the effect of a change in plain English, so that the consequence of a weighting decision is legible to a non-statistician.
Freedom is double-edged. The same adjustability that makes the model honest also lets a user engineer a ranking to fit a predetermined conclusion. The uncertainty flags are a partial guard, but the tool assumes good faith.
Illustrative zones. The configurator runs on a set of realistic Kent PCN zones for demonstration; the production heat map runs on the full live district data. The configurator teaches method; it is not the operational scoreboard.
Weights are not effect sizes. A 25% weight does not mean the signal causes 25% of frailty. It is an analyst's statement of relative importance, not an epidemiological coefficient.