Four Hundred Parts per Million Is Only the Beginning: Three Cheap CO2 Monitors Through an Autumn Term
A classroom of 32 Year 8 pupils will push carbon dioxide past 1,500 parts per million within half an hour of the first bell if the windows stay shut. Outdoor air in a British city sits at roughly 420 ppm; human breath does the rest. Since the post-pandemic rush to improve ventilation, the cheap NDIR monitor has become the default instrument for judging whether a room is adequately aired, and it is now as likely to be found on a school business manager's requisition list as a thermometer.
The question CavLab set out to answer was narrower than whether these devices are useful. It was whether the number on the display can be trusted when someone is standing in front of it deciding whether to open a window.
Three monitors and a reference
We instrumented a single mixed-ability classroom in a 1970s Hertfordshire secondary school for the first nine weeks of the autumn term, from early September to the start of November 2026. Thirty-two pupils, one teacher, two external walls, four opening windows and a door onto a corridor. No mechanical ventilation beyond extract in the adjacent toilets.
Three monitors were bought at retail, all advertising non-dispersive infrared sensing rather than the cheaper and considerably worse metal-oxide approach:
| Unit | Price | Claimed accuracy | Sensing | Display |
|---|---|---|---|---|
| Meter A | £89 | ±50 ppm ±5% | NDIR, single beam | 3-digit ppm, traffic light |
| Meter B | £64 | ±(50 ppm + 5%) | NDIR, single beam, ABC | 3-digit ppm, icon only |
| Meter C | £99 | ±(30 ppm + 3%) | NDIR, dual wavelength | ppm, graph, logging |
All three were compared against a loaned reference analyser with a calibration certificate dated August 2026, specified at ±(50 ppm + 5% of reading) and logged at one-minute intervals alongside the monitors. Temperature and relative humidity were recorded independently. The monitors were sited at 1.2 metres, away from windows, radiators and the teacher's desk, and rotated between three positions each fortnight to separate siting effects from instrument effects.
The baseline that quietly walks
The most consequential finding was not accuracy at high concentrations. It was behaviour at low ones.
Two of the three units rely on automatic baseline calibration, an algorithm that assumes the sensor will periodically see fresh outdoor air and uses that as its zero. In a domestic living room this is broadly reasonable. In a school in November, with windows shut overnight and a caretaking team locking up at six, the room may not fall below 700 ppm for days. The algorithm, given no true reference, adjusts the baseline upward. The sensor then under-reports everything above it.
By week nine, Meter B read 918 ppm when the reference showed 1,050 ppm during occupied lessons — an error of 132 ppm in the reassuring direction. Taking it outside on a still Sunday morning, it settled at 468 ppm against a true 421 ppm. Meter A drifted more slowly, ending the period about 38 ppm low at the 1,000 ppm mark. Meter C, which uses a dual-wavelength arrangement and a manual calibration routine, held within 12 ppm of the reference throughout, and its own calibration prompt was the only one of the three that ever appeared.
This matters because the thresholds schools actually use are close together. A reading of 1,500 ppm on a drifting sensor may represent closer to 1,700 ppm in the room, and the difference between a room that needs a purge and one that merely needs a window on the latch is exactly the sort of margin a cheap instrument erodes.
Humidity, cold draughts and the problem of siting
We had expected condensation to be the main confounder. It was not. Relative humidity in the room stayed between 38% and 66% for most of the term, and all three units tracked within specification across that band. Meter B became erratic above 72% RH during a wet week in October, over-reading by up to 90 ppm in short bursts; the other two were unbothered.
Temperature was a smaller factor than siting. A monitor placed on a windowsill read 300 to 500 ppm below the room mean whenever a window was open, because it was sampling the incoming cold air rather than the occupants' air. A monitor at 0.8 metres, at seated head height, read noticeably higher than one at 1.8 metres. Neither is wrong; they are measuring different things, and a display that offers one number conceals that.
The response time also deserves scrutiny. Stepping from 600 ppm to 1,200 ppm, Meter C reached 90% of the change in 45 seconds, Meter A in about a minute, and Meter B in nearly two. For a device meant to prompt a window-opening decision, a two-minute lag is tolerable. For a device being waved around a room to find a stale corner, it is not.
Reading the trend, not the number
None of this makes the monitors useless. It makes them instruments with a known failure mode, which is a different and more workable proposition.
A monitor that under-reads by 100 ppm will still show the characteristic sawtooth of a room filling during a lesson and emptying at break. It will still reveal that the back-left corner is worse than the front. It will still show that a door propped open with a wedge does more than two tilted windows. As a relative instrument, all three earned their place.
As an absolute one, only Meter C behaved in a way we would be comfortable defending, and it cost roughly what a school spends on a set of atlases. The cheaper units are not dishonest, but they are not measuring what their three-digit displays imply.
What a school should actually do
Four practical points emerged from the term.
First, check the baseline. On a calm day, take the monitor outside for ten minutes. A unit reading above 500 ppm in fresh air has a baseline problem, and every subsequent reading will be low.
Second, do not rely on a single display for a decision. Two monitors in the same room, disagreeing by more than 150 ppm, tell you something useful about both of them.
Third, log rather than glance. The traffic-light icon encourages a binary judgement at a moment in time; a week of one-minute data shows the pattern, the recovery time after break, and whether the ventilation strategy is actually working.
Fourth, treat the number as a proxy for air freshness, which is what it is, and not as a compliance measurement. Where a school needs to demonstrate ventilation performance, a calibrated instrument and a documented survey remain the appropriate route. Cheap monitors are a prompt, not a certificate.
The limits of one classroom
This was one room, in one school, over nine weeks. It is not a survey, and we would not present it as one. The reference analyser itself carries an uncertainty of roughly 100 ppm at the concentrations we cared about, which is the same order as the errors we are describing — a point worth holding onto before anyone draws a firm ranking from our table.
What we can say is that the instruments agreed with each other more closely than any of them agreed with the reference, and that their disagreements from the reference leaned consistently in one direction. A monitor that fails by flattering the room is a worse instrument than one that fails noisily. In a British classroom in November, that distinction is the whole argument.
