That’s because the table includes the (non-scaled) M value at limitJMax. Should it not be possible to derive the 100 nits equivalent reach M value from a 1000 nits (or 10000 nits) equivalent reach M? Edit: I haven’t tested the following but I think it is:
M_{100nits} == J_{100nits}^{\frac{1}{cz}}* M_{10000nits}
And that is what happens, in Blink as well. It’s two different lookups, though. I agree it’s a bit strange, but we never went on the journey making all the tables have same spacing.
But we don’t use D65 in reach mode. The reach mode operates always, in gamut mapper as well, with ACES white point (as it’s AP1). If the white point of the reach table is changed it won’t invert exactly to AP1 boundary (we had that bug not long ago).