typo

should this be "are not randomly distributed"

? typo?

2 typos here.

this might be true for some devices, but not all devices...

I recognized that this is offered as a simple and cost effective method, but there are significant risks in focusing methods that focus on 1 light source and/or 1 intensity range, as different devices might perform great under some sources and terrible under others (e.g. sources that are included in their training dataset vs sources not included) and similarly, some might be great at some intensity ranges but less great outside that range.

Some devices allow for networking or have APIs that allow for additional capabilities such as real time monitoring, integration to broader data acquisition systems, or integrations with lighting control systems.

our new device charges wirelessly via a Qi charger

This is an important point that I wonder if could be future emphasized graphically. Is there something from the references that demonstrate this? Or I think you might be doing a study now with sensors with multiple wearable locations that might allow you to make this point graphically with real data.

Does it make sense to discuss sleep tracking data here at all? We pull in sleep data via our app from Apple Watch users using a HealthKit integration and I believe that other manufacturers also measure (or integrate) sleep data as well.

We calibrate each channel of each of our sensors using a complex process that involves training data from 1000's of light different light sources. We don't recommend that users calibrate our devices themselves, in part because don't believe they would have the information needed (e.g. training data, channel response characteristics, channel gain and integration times, etc) to calibrate them to measure SPD, mEDI, lux, etc as accurately as we are able to.

Why not focus on spectral power distribution (SPD) instead of sensor channels? Accurate, absolute SPD values are what we really care about when characterizing the spectral environment whereas channel responses are inherently technology dependent. I see more risk for confusion and misapplication from using channel data rather than SPD results. I expect this to become even more pronounced as manufacturers like us utilize AI/ML calibrations methods, which are likely to further improve SPD accuracy but do so in ways that are less clear to researchers on how channel responses correlate with SPD values.

UV and IR are discussed later in the "Advanced and additional modalities" later but it could be useful to discuss here (or note that light outside 380-780nm are discussed later), especially as interest seems to be growing on the impact of UV (METROLOGY FOR NEUROPIC RESPONSE TO LIGHT: A CASE STUDY IN A NEONATAL INTENSIVE CARE UNIT by Mehlika INANICI, James M GREENBERG, Richard A LANG, Jaime STRUVE, Katherine A GRUNER. ) as well as in the IR (inflammation, mitochondria health).

I agree that sometimes selecting the device is the right first step, but I would argue that in many cases, it is better to start with defining the study goals and protocols. As discussed below, there are a number of tradeoffs with devices (size, accuracy, wear location, user-friendliness, battery life, etc) and it can be useful to clearly define the research objectives before selecting the device so that the device(s) used are a good fit for the study. Starting by selecting a device may limit the ability of researchers to design their protocol gather the most relevant data to meet their researcher objectives - even if other devices are available that can gather these data but have been screened out in step 1.