We invented a government cover-up that does not exist. Here is how the AQI in your phone is actually calculated.

In June we published a piece on this page describing a government smoothing mechanism we called the 'Variance Protocol', and said it hides the worst hours of an air-quality day inside a 24-hour average. There is no such thing as a Variance Protocol. We invented the term. And the mechanism that does exist was built to do close to the opposite.

What the EPA actually does

Two different numbers get called the AQI, and almost every argument about air-quality data is really an argument about which one somebody is looking at.

The DAILY AQI is what the name suggests: a 24-hour figure for PM2.5, reported for the day. It is the number in a forecast and in the record. A day with three hazardous hours and twenty-one clean ones can indeed come out Moderate, and that is arithmetic, not concealment.

The number in your phone is usually the other one. Real-time reporting on AirNow uses the NowCast, and the EPA's technical assistance document sets out exactly how it is computed. The PM NowCast is an average of the previous twelve hours — not twenty-four — and the weighting changes with the air. When conditions are stable the twelve hours are weighted roughly evenly. When conditions are variable, the most recent hours are weighted far more heavily, down to a floor the document describes as approximating a three-hour average.

The weighting is not a judgement call. It is computed from the data itself: take the highest and lowest of the last twelve hourly readings, divide the range by the maximum to get a rate of change, subtract that from one, and floor the result at 0.5. Fast-changing air produces a small weight factor and a short effective window. Still air produces a large one and a long window.

Why it was built that way

The EPA updated the PM2.5 method on 1 August 2013 and the PM10 method on 9 December 2014, and the document states the purpose plainly: it was designed to be more responsive than the previous method in rapidly changing air quality conditions, such as those we see during fire events. The ozone method was updated on 1 August 2019 on the same principle.

In other words, the specific failure our article described — a smoke plume arriving at four in the afternoon and vanishing into a day-long average — is the failure the current method was written to prevent. It is not perfect, and it is still an average, but it is an average that shortens itself when the air starts moving.

What is still worth watching

Three things survive the correction, and they are the part of the original piece that was worth writing.

The daily figure really is a 24-hour average, so a 'Moderate' day genuinely can contain hours that were not. If you are deciding whether to run this afternoon, the daily number is the wrong one to read.

The NowCast needs data to work: the document requires that two of the last three hours be valid before it will compute at all. A monitor that drops out during the worst of a fire produces no number rather than a bad one, which is the right failure but an easy one to misread as clean air.

And a monitor measures where the monitor is. Sacramento County's reference instruments are a handful of points across a basin that traps smoke unevenly against the Sierra. Del Paso Heights and Land Park can differ by more than the difference between two AQI categories, and no averaging window fixes that.

Where to check

AirNow publishes the hourly concentrations the NowCast is computed from, and the EPA's concentration-to-AQI converter will turn a raw microgram figure into a category. If a number ever looks wrong to you, those two together will settle it in about a minute — which is the answer our first version should have pointed at instead of inventing a mechanism to be angry about.