How these numbers are made

The headline figure is the CPCB National AQI, computed from 24-hour mean concentrations — which is how CPCB defines it. Not the latest single hour, which moves far more and reads higher.

Readings come from the nearest monitoring station, and the card always says how far away that station is. India's regulatory network has been silent since 27 August 2026, so most localities are currently served by community low-cost sensors measuring PM2.5 alone. Those cards say exactly that rather than presenting the number as a full AQI.

School locations come from OpenStreetMap, staffing and enrolment from UDISE. The two disagree constantly — a locality can have sixty mapped schools and staffing figures for one — so the card reports both numbers rather than the flattering one.

The UDISE snapshot is from January 2022. That is disclosed on every card and is why schools never reaches high confidence. UDISE carries no exam results at all, so nothing here is a ranking of school quality.

Both come from local news coverage, read and classified per locality. They are the two categories where official Indian data is weakest: NCRB publishes crime at district level, a year or more late.

Neither is ever turned into a score. How often a place appears in the press tracks how much media attention it gets, not how often things happen there — a well-covered neighbourhood would look dangerous next to an identical one nobody writes about. So we describe the kind of incident reported, which survives that bias, and leave the ranking alone.

Every headline is judged individually for whether it describes an incident in that specific locality. The card shows how many were read and how many survived, so the count never reads as an exhaustive tally.

A weighted composite of the categories that can currently be scored, renormalised over what is actually present. Missing data never counts as a bad result — a locality with no air quality reading is not scored as though its air were poor.

The score names what it is based on, because it currently rests on two of six categories. Below that threshold no number is published at all, and the page shows the individual cards instead. A single measurement wearing the words “Trust Score” would be worse than no score.

Every figure carries one, and it is decided at the moment you read the page rather than when the data was stored — so a reading that was fresh when fetched degrades on its own as it ages, and is withheld entirely once too old to be meaningful.

Community-estimated means there is no reliable official source and the figure leans on press coverage or community sensors. It is never dressed up as official data.

• Publish a safety score from press coverage. The bias runs the wrong way and we have no way to correct for it yet.
• Report a median from too few samples, or a school count we can see is wrong.
• Fill an empty category with an estimate. Power and infrastructure have no data, and their cards say so.
• Average away a disagreement between sources. Where two sources conflict, the report shows both and says which is which.

Some of this will be wrong. A locality's name can appear in a national story that has nothing to do with living there; a monitoring station can be further away than it looks; a 2022 staffing figure can describe a school that has since changed completely.

Where we find such a case we fix it and say what it was, rather than quietly adjusting the number. If you spot one, tell us — you probably know your own neighbourhood better than any of these sources do.