After a long Odyssee, I finally figured it out. - I could not understand why scanin did not recognise the tif after I put scanin in the same folder. I suspected scanin was corrupt, and this seems to have been the case, since installing the latest version of Argyll fixed it.
However, the resulting profile had a peak error of 78.996748 and an average error of 23.109304. After much wondering, the diag tif gave a hint: The tif was downscaled to 1/4 the size, and Argyll placed the frames (or what they're called) partly outside the actual color patches.
OTOH, using the full scale tif resulted in error "pattern match not good enough", as exspected, which is because the noise of the black frame breaks up the pattern. So downsizing the input tif to 1/2 was the proper compromise. Maybe it also helped to lower the resolution from 360 to 72 dpi.
The resulting profile has a peak error of 8.118006 and an average error of 2.438826. That's acceptable.
AlterEgo, in case it has your interest for your own profiles: MakeInputICC displayed a neutrality check for the profile it made from the raw. The graph shows a divergence between the R, G and B curves, regardless if I chose White Point unscaled or auto-scaled.
Opening the profile in ColorSync and looking at the R, G and B response 'curves', they all have slightly different gamma, and all slightly different from 1.
I'm afraid this would translate to a red cast in a real image.
Braggin' ;-) : The profile I finally managed to produce has gamma=1.0 for all three channels.
I used the following options:
scanin: -G1.0: approximate gamma encoding of image
colprof: -am: creating a matrix profile with gamma=1
-u: If input profile, auto scale WP to allow extrapolation; this means the profile can reproduce whites which are whiter than the whitest patch in the target, here the ColorChecker. This is unlike the standard behavior of profiling.
Furthermore, I manipulated the .ti3 file to allow for the production of pure black:
I added a line 'NEU00' with all values 00.0000, and accordingly increased the number of data sets from 50 to 51. (Tip from Brian Griffith, author of Iridient Developer)
MakeInputICC does not give me all these (and other) choices. There is a price for the convenience of a GUI...
Thanks to you who chimed in!