Methodology

The reading

What a number means

Every city carries a 0–10 reading per vertical. A reading is the mean of that vertical’s stored values for the city, normalised across the observed range of all cities holding data for that vertical — a min–max scaling, because the underlying metrics do not share a native scale. Sentiment is scored out of ten; developer activity is an index; property appreciation is a percentage. Comparing them without normalising would be arithmetic on incompatible units.

Weighting

Coverage is the weight

The living index divides a city’s summed normalised verticals by twelve — not by the number it happens to hold data for. A city with three covered verticals therefore cannot outrank a city with twelve on thin evidence. That division is the coverage weighting; there is no separate multiplier.

Provenance

Observed versus modelled

A figure is observed when it came from a live source in this run: OpenStreetMap Overpass, Ticketmaster Discovery, Open‑Meteo, GDELT, the World Bank, WAQI, Transitland, OurAirports, or a human check‑in. A figure is modelled when its shape or magnitude was derived rather than measured — for example a time‑of‑day curve with no measured hourly source behind it, or a seat‑capacity figure derived from airport class. Modelled figures carry a MODELLED label in the interface. They are never presented as measurement.

Airport data is always described as scheduled seat capacity. It is never described as arrivals and never as passengers, because OurAirports publishes infrastructure, not traffic.

Thresholds

When we publish nothing

A vertical publishes for a city only once it clears a minimum of two stored signals. A Tonight mode publishes only when a live source returned something for it. Below those thresholds the interface shows a designed empty state. It does not show a spinner, and it does not show a number, because an invented number is worse than an honest absence.

Confidence

The ramp

Each stored reading carries a confidence between 0 and 1 set by its source. Readings below 0.6 are treated as modelled for the observed‑versus‑modelled split. Confidence rises with signal volume and with independent corroboration across sources; a single check‑in moves a city’s reading very little by design.