According to the Unihedron website the SQM are calibrated to a NIST standard. But if you dig a bit further you'll find the "NIST standard" is basically a commercially made photographic light meter, designed to measure in tens to hundreds and thousands of lumens. That is -several orders of magnitude brighter than anything the SQM is intended for. I don't see how a sensor calibrated to those levels so far off can be called 'calibrated' for very tiny levels of light. The sensor and electronics would have to be extremely linear in the response over many magnitudes of brightness. Also, the sensor is not sensitive to blue wavelengths, which is where the vast majority of lumen output of LED streetlights falls, along with increased scattering of bluer light, the SQM will 'not see' much of contribution as more and more LEDs contribute to light pollution and so underestimate the amount of light pollution. It may even lead to the wrong conclusion that light pollution is decreasing over time. An example of bad data is worse than no data.
I see many discussions on CNs of observers 'splitting hairs' in arguments about dark skies and tenths of readings completely unaware of the difference between precision and accuracy v. full scale calibration range of instrument readings. A SQM precisely fits the old engineering & science adage of "measure it with a micrometer, mark it with a chalk, cut it with an axe."
And yet it works. So the assumptions above should be evaluated.
- I don't know that the sensor and electronics have to be that linear. Afterall, the magnitude scale is not linear, but logarithmic, and this serves to smooth the errors. In a 2005 Italian evaluation (link) the resultant magnitude standard error was +/- 0.028 mag over a range of about 8 to 20 MPSAS...and it actually had lower deviation at both ends of that range than in the middle.
- The spectral response of the meter appears to be pretty strong throughout the relevant visual scotopic wavelengths, including the blue, still about 75% of peak at 450nm, and somewhere in the 90's at 510nm. So it is going to be able to detect the LED emissions with only somewhat reduced efficiency rather than "not see much."
- Your argument is perhaps backwards...the SQM is also sensitive to the red that we don't see well at night, so the lack of red contribution makes it harder to correlate to some existing photometry systems that are heavily weight toward the yellow or red when the sources change to a bluer type. The V band and CIE photopic band have the same problem, even more so. The CIE scotopic should be somewhat better in that regard (being insensitive to red), but ironically it doesn't capture the blue as well as an SQM will.
- The closer conditions are to pristine, the smaller any error introduced by LED or other manmade lighting sources. That is because the fraction of the manmade sources shrinks so much. So counter to the argument you have made above, darker conditions will show the least departure for visual. The actual concern would be for light polluted conditions.
When it comes to light polluted conditions I have seen some interesting things scotopic-vision wise. In an area of rapid suburban sprawl around me, I noticed the sky becoming slightly darker with conversion to LED streetlamps, despite being the less desirable 4K variety. Being "full cut off" they throw quite a bit less light into the sky, and I can see that. I can also see how they throw an order of magnitude less light into my bedroom at night (light replaced this week.) The HPS lamps stuck out and threw a great deal of light to the side and up. I first noticed the improvement at the eyepiece when the neighborhood next door was converted about 2 years ago. And the SQM-L indeed showed some improvement as well. I expect this to be short lived with the new homes and schools being built, etc., but it ran counter to the conventional "wisdom" about LED street lights. And if these were warmer LED's the concern would largely disappear.
No single meter can be trusted for anything if one does not have ways to cross check it. The SQM is no different than many other meters in the garage or elsewhere in this regard.
The ones discussing these things tend to know a lot more about calibrating and measuring things than you give us credit for. I will take a meter that gives reproducible results (and can be tied to various calibrations) over somebody's +/- half a magnitude approximation with uncalibrated eyes any
day night of the week. Even if the meter is shown to have a significant departure from actual, that can be corrected for accuracy and regardless it still provides an orders of magnitude more precise value for comparison on its own.