Originally posted by Eruditass great, detailed, and thorough tests. I personally have avoided commenting because I don't completely understand all of it yet! especially how ISO100 doesn't have ~1 stop less of DR than ISO200.
Theoretically, going from ISO200 to ISO100 should increase DR by +0.5stops (not decrease by -1stop). ISO100 captures twice as many photons which results in 0.707x the noise.
Because the K-x hasn't native ISO100, it should make no difference though. And indeed, the difference for gray is almost null.
But digging into my data, I see that in a log-log plot, the SNR curves at ISO 100 and ISO 200 between 1% and 100% are both linear and meet each other at the 100% end. But slopes are different and at the 1% end, the ISO100 curve is 3 dB apart like for a native ISO 100 sensor.
The ISO 100 curve is flatter than pure photon shot noise would allow.
Currently, I don't know how they did it
To verify that I have no stupid measurement errors, I today had a lokk again and appended to my K-x sensor article wrt to the "ISO 100 bump".