Very good questions! While those graphs are very typical to interference filters, there are indeed some points of concern as you pointed out.

I used a successor of Hamamatsu diode S12915-1010R  (https://www.hamamatsu.com/eu/en/product/type/S12915-1010R/index.html). While Hamamatsu specifies that this diode is sensitive up to 1100 nm, it still works till theoretical Si diode limit at 1200 nm as it should be - so in that region detector is not exactly super sensitive. Each datapoint on my graphs is an average of 10 individual readings (i.e. readings are definitely not high-frequency temporal noise) and diode setup was fed by an alkaline battery. I had to scan one of the filters more than once and it's NIR behavior did not change over several hours, that gives me the confidence to say those peaks are real. Of course, it is stretching the limits past 1000 nm, but I was pretty confident about the quality and stability of equipment. In addition, measurements were not in photon-starving regime around 1 µm!

The setup was:

1 kW halogen bulb -> condenser lens -> double monochromator -> aperture -> projection lens -> filters in the filter wheel -> silicon photodiode catching all the light from monochromator, everything collimated by a laser.

All curves were collected by scanning with monochromator 3 nm step by step from 350 nm to 1200 nm. Every scan was forked by a reference scans (without filter) with exactly the same parameters. Of course, everything starting from the exit opening of the monochromator (included) was in a completely light-tight box. I followed standard procedures in our radiometry lab.

I suspect that I can't properly answer your question about IR scattering. Probably changing slightly the tilt of filter and re-measuring eveything would give slightly different results when there is some kind of scattering happening. And it seems, that I really have to repeat measurements with an InGaAs detector, too!

I haven't been too worried about those very far NIR leaks, because typical CCD QE there is very low and my typical targets are not very red ones. So the that leak really could be detected when the object is an extremely red one (V838 Mon?). If I recall correctly, Arne has done such test before and he did not found anything problematic. Regarding atmospheric effects past 1 micrometer - unfortunately that area (1-1.2 µm) is just moderately attenuated by atmospheric absorbtion (see e.g.: http://www.gemini.edu/sciops/ObsProcess/obsConstraints/transnir1.gif).

Still, maybe there is some kind of lab setup that could be used as well (ordinary incandescent light with and without NIR blocking filter) having both filters and the camera in test. I will discuss my colleagues about that.

By the way, technically it is possible to extend that NIR blocking range further into the red. Few years ago I iterated (with Schott) a design of a custom NIR blocking filter and every attempt to ensure wider blocking range resulted in (a bit) worse behavior (typically more waving and less transparency) in visible light and quite a bit of increase of the cost. IMHO Astrodon found a sweet spot. :-)

Best wishes, Tõnis

Update. Good point from our lab engineer about scattering was: scattered light from surfaces is almost always faint in such setup and because only monochromatic light exits monochromator, the contribution from scattered light (it's intensity x very low QE of sensor) is very low, next to zero.