How a reading gets made

Methodology

The plain version

How a reading is computed

Every observation - a submitted reading, a GDELT sentiment pull, a World Bank indicator, an air-quality reading - lands in one append-only table, tagged with a city, a vertical, a value, a source, and a confidence score. A city's current reading for a vertical is the confidence-weighted average of everything recorded for it in the trailing window, not a single latest value. Nothing is ever overwritten; old readings age out of the average, they aren't deleted.

How inputs are weighted

Two things move an input's weight: how confident the source is in it, and how recent it is. A same-hour GDELT pull carries more weight than a three-day-old one. A reading built from fifty inputs carries more weight than one built from a single input - see the ramp above.

Why an early reading counts less than a later one

A single observation is real, but it's one data point in a place that never stops changing. The confidence formula treats sample size as the thing that converts an early observation into a trustworthy reading - not because the first input is wrong, but because a broader sample is harder to be wrong than a narrow one.