I assume the underlying blue emitters in different CCTs or CRIs of the same LED would be the same, so it comes down to phosphor conversion efficiency. Conversion should not be 100% efficient, so more conversion (lower CCT or higher CRI) should be less efficient.
For the question of how much of the lumen loss of SFT-40 3000K is due to efficiency loss, I guess it would be trivial to calculate by applying the luminosity function to the emission spectra, but it is a pain to do so without the raw data.
Many important computations should be trivial, but for whatever reason it is extremely difficult to find data in a usable format. I havenāt been able to find any usable data for the luminosity function, or for the relative spectral power of a test emitter.
I was able to find some luminous efficiency tables, but for emitter spectral power distributions youāll have to ask LED testers. Going down the rabbit hole, I found there are several photopic luminous efficiency functions. Here is the CIE spectral luminous efficiency table for photopic vision. There is also the Stockman & Sharpe version. The differences are quite large; you can visualize them in this Wikipedia plot.
At 450nm (which LEDs usually peak at), the CIE version says 0.038, while Stockman & Sharpe says 0.0647. This is 70% higher and will greatly change lumen numbers depending on which you use. CIE should be more frequently used for measurements but Stockman & Sharpe might be more accurate.
Moreover, when using flashlights in the dark, we likely engage mesopic vision instead of photopic, which introduces even more complexities. Therefore, lumens are not an accurate measure of how bright flashlights are. It would be easier to directly measure and compare radiant flux and forget about lumens altogether.
A better measurement than lumens or lux might be a hypothetical detail resolution unit. Pure green has the highest lumens or lux but provides low detail resolution. LEDs with strong blue peaks also have lower detail resolution because the fovea is dominated by red and green cone cells. The luminous efficiency function has undoubtedly contributed to the prolonged era of cool white, green tint and low CRI LED lighting, which doesnāt allow people to see clearly but has more lumens.
Thank you very, very much for doing the research and finding this data!
It is frustrating that the luminosity function is not a real physical quantity (i.e, derivable from expressions solely in terms of physical constants), and is thus very difficult to sensibly interpret, not to mention to pin down exactly. The existence of two versions testifies to this. Fortunately, the disagreement seems to happen mostly at the left tail, so for neutral-warm CCTs one can expect a fairly small difference.
Very good point about mesopic vision, which distorts relative brightness perception and therefore both perceived brightness and tint. I do find radiometric power a much, much easier quantity to work with compared to lumen; plus, one cannot cheat this metric via excessively green tint.
Your idea of detail resolution unit is very interesting, I need to think more on it. Lots of folks default to recommending the Osram green when asked for a throwy LED, which is really a bad idea as an unqualified recommendation. I also notice lower detail resolution with very cold or blue LEDs; pinpoint sources of monochromatic blue always seem blurry no matter how hard I try to focus, and a cool CCT in low intensity seems to cast a grey veil over everything.
I, for one, am looking forward to reviewing the sft90 in the new Fenix TK21R, as has been arranged for me recently. Iāve been so dissappointed with SBT90.2 since its first appearance due to the heat sag and less-than-desirable tint shift.
