I don't, unfortunately. I did plan on publishing something about it, but life got in the way. Maybe you'll see me post something about it on HN one day.
In summary, there's nothing really groundbreaking about what I did other than I made it smaller and more portable than most existing indirect calorimeters. I made a circuit board with some sensors and an Arduino Nano mounted on it. The outer shell was designed with OpenSCAD and 3D printed. It was designed so it could be worn on a facepiece (in my case, a modified 3M respirator).
> Why is the indirect method so sensitive to extraneous movement?
Anything movement made is a result of metabolic activity. I was surprised to see drastic changes in RQ (respiratory quotient) just by getting up out of my chair and walking to the bathroom. It can take time for reading to stabilize. One reason is that even the best CO2 sensors have a slow response time in contrast to O2 sensors. There's enough lag that a change in activity can ruin a large section of a test, in particular if you're anticipating delayed metabolic activity. Also, it takes the body some time to eliminate CO2 after any amount of exercise. After movement, especially something like steady state cardio, this causes the RQ to jump up for ~3 to 5 minutes before it drops down.
One thing that research grade ICs do to mitigate this is to use a mixing chamber with a sampling pump to try and smooth out and normalize readings over a window of time. My approach was to literally just have my breath blow over the sensors with valves only allowing air to move in one direction, which is simpler and allows for readings to be a bit closer to real-time. It's also considered more problematic than other approaches like the mixing chamber.
Oh yeah, there's also this thing with lactic acid buffering that can cause some extra CO2 production but isn't necessarily considered metabolic activity.
Then there's the problem of leaks in the system, which are more likely to occur when the subject is moving. Even a little air leak can create anomalies, and you don't always know when they occur.
There's a lot of confounding factors, and I'm sure I'm forgetting some. Unless an indirect calorimeter has been designed by a company specifically for variable movement, you can assume that the only way to get reliably results is to make sure that the metabolic activity remains consistent for the duration of a test. That means either the subject lies still and doesn't move at all or they're performing something like cardio at a steady pace. If you look up protocols for conducting IC, you'll notice they're very strict.
> More so than just reflecting the additional energy expenditure?
Energy expenditure is thrown off but actually much less so because it is more closely tied to oxygen consumption than RQ, which is more closely related to the volume of CO2 produced. Since oxygen sensors respond fast and the body doesn't do weird things with oxygen like buffer it, EE isn't affected as badly. But if you want to measure how much carbohydrate to fat is being utilized, then any disruption can cause confusing results.
Indirect Calorimetry is very difficult to get right, but it's used because the alternative, direct calorimetry, is usually impractical. Direct calorimetry of a human being involves placing the subject in a room with a water jacket and measuring the difference in temperature after the subject has radiated heat away from their body. It avoids the confounding factors of IC, but you can't measure RQ that way and it's not really practical as I've said outside of financed research.
Why is the indirect method so sensitive to extraneous movement? More so than just reflecting the additional energy expenditure?