The bedroom cardio detector had been quiet and reliable for weeks, then logged two sessions nobody had. Both short, four and five minutes, both at bedtime. The household's report was precise: "we just went to sleep". Bedtime, it turns out, has the same signature as exercise, and the detector's most clever branch was the one that fell for it.
The bedroom cardio detector is one of the flat's more delicate instruments, and for weeks it behaved. Then it logged two sessions nobody had, both at bedtime, both short, and the household's report was admirably precise: 'we just went to sleep'. Going to sleep, it turns out, looks a great deal like exercise from the point of view of a CO₂ sensor.
Two bodies, one signature
The detector's main branch is strict: a steep CO₂ rise and a humidity rise in the same room. That branch was innocent. The false alarms came from a second branch added in July for the draughty summer nights when the whole flat's humidity rises together and the bedroom's relative reading tells you nothing.
That branch trusts body heat instead: a slow rise in the temperature near the pillow, against a colder outdoor draught, plus the CO₂ trend, plus somebody present. Only bodies add heat to a room that is venting cool air, the reasoning went. True. Two bodies getting under a blanket add heat too.
The night's data made it plain. The room went from 24.2 to 24.8 degrees in twelve minutes; CO₂ climbed at twelve to fifteen ppm a minute, exactly at the trend's threshold; humidity did not move (+0.3 percent). A real session the week before: CO₂ at 38 ppm a minute and humidity up 2.8 percent in a quarter of an hour. Different magnitudes, but a threshold-based trend cannot see magnitude, only "rising".
Persistence is the discriminator
What it can see is time. Going to bed pushes CO₂ up for about four minutes, until the open window and the breathing find their balance and the trend flag drops. A real session keeps it rising for eleven minutes or more, and the shortest genuine one on record, ten minutes, still clears it.
So the draught branch now demands that the CO₂ trend has been on for at least five minutes before it will believe body heat, and a trigger re-evaluates the detector at the five-minute mark. The genuine session from the week before would have been detected at the same minute as it actually was. The two false ones would not have fired at all.
The bookkeeping matters as much as the fix: the two phantom sessions were removed from the permanent log, the lifetime counter went from twelve to ten, and the nightly reviewer now knows that a "cardio" flag appearing within minutes of the bedroom radar turning on and lasting under eight is a probable false positive worth reporting.
The detector is quieter now. So, we are told, is bedtime.