This is my last try - I mean it this time!
Yesterday I tried a different technique for selecting filters to get the PAR correction on the Mastech sensor. I tried to visualize the appearance of the sensitivity plot needed to make the change to the sensor sensitivity I wanted. Then I searched until I found one that looked like it should, and calculated the sensitivity with that filter in place. Then I just repeated this process until I was near the amount of attenuation of the output that I needed, and assembled the sensor with those filters to see how close I was. That let me calculate the % transmission I needed in one last filter. After several tries I got the spectral sensitivity near what I wanted, and the readout very near what I wanted. I ended up using 6 filters - Roscolux #17, 36, 55, 363, 4430, and 3304.
Two more discoveries: Most important, the conversion from lux to PAR for this meter isn't dividing lux by 61, as I have been saying, but it is dividing lux by 76 - big mistake on my part. This is the reason the data we got for CFL bulbs was so far off from the older data. Then I find that the Mastech sensor electronics are easily confused. Every time I open the sensor case and replace it, the readout goes crazy for a few minutes, before settling down to read correctly. This is probably because of the cheap way it is made.
The sensitivity chart for this final sensor looks like:
When I check this against the Quantum PAR meter I get 32 PAR with the Mastech meter and 29 with the Quantum meter. In the sunlight, the Mastech meter reads appropriately. (I wasn't looking for exact readings for that much light)
I found the easiest way to add these filters to the Mastech sensor is to just lay the stack of .5 inch square filters on top of the circular housing for the diode, and use a couple of strips of Scotch tape across opposite corners to hold them in place. That part is extremely easy. If a correction is needed it is easy to remove the stack and substitute for one of the .5 inch squares.
When you stack filters it tends to result in exaggerated peaks and valleys in the response curve. First, the sensitivity is cut by a factor of 76, and the filters that are appropriate tend to have peaks and valleys in roughly the same places, which quickly amplifies the magnitude of those peaks and valleys. One would think that neutral gray filters wouldn't have those peaks and valleys, but they do, and the darker they are, the bigger the peaks and valleys. Filters that are the darkest also tend to totally block certain parts of the spectra, which isn't good at all, so multiple different filters are necessary.