Not sure about the fun bit but it's definitely necessary. Because I have the luxury of being able to set the temperature of the sensor to -25°C (my chosen working figure) at least I can do the calibration frame exposures with the camera indoors.
RBI definitely exists. As an illustration here are three dark frames straight out of the camera with no processing other than exactly the same stretch applied to all three:
The three dark frames were all 200 second exposures and the first one was taken without the camera's built-in infra-red LED having been used for a long while. The one on the right shows the effect of using the LED to saturate the sensor for 45 seconds followed by a discarded very brief "dark" exposure and then the 200 second dark exposure you can see. The middle image was taken after the 45 second flood, a brief dark exposure, a 10 second dark exposure and a 100 second dark exposure, all three of which were discarded before the 200 second dark exposure you can see. Hope that makes sense!
The 100 second dark pre-exposure helps bleed a little of the RBI signal but not a huge amount. That said every little helps (see the average signal deviations mentioned in the next paragraph) so I intend to stick with it for all my exposures whether or not I use the LED to flood the sensor. At first glance the immediate conclusion to draw is not to use the RBI flood LED at all but the point is that if the previous exposure contained bright areas then the next exposure will, if long enough, also include an echo from the previous exposure. With perfect tracking on the same subject that might actually be considered a win but tracking is rarely perfect and in any event I prefer to dither the exposures by moving the pointing a few pixels between frames as that helps the stacking algorithm reject bad pixels that haven't already been mapped out by the calibration frames. The bottom line is that with a cooled CCD RBI is something that has to be recognised. Uncooled sensors don't really see this effect but then they see a lot
more noise from other sources.
It's easy to look at the illustration above and believe that the IR LED is causing a lot of background signal but that's because of the way I've applied the stretch. The median signal from the images, working from left to right, is 1994, 2037 and 2045 (16 bit readout) with corresponding average deviations of 7.3, 8.9 and 9.9. Median signals can always be subtracted so the difference isn't huge - it just looks huge in the illustration because the stretch is so close to the noise floor. As an aside, both of my recent Hα images were taken with the RBI flood enabled and I was able to extract plenty of the faint stuff before the banding you can see became an issue. As I understand it that banding is a result of the way the chip is made rather than uneven IR flood illumination. There is a rather more obvious chip defect if you look very closely as one half of one column is completely dead. Not unusual, I believe, for such large sensors unless you pay considerably more and it's another reason why dithering the pointing between exposures is a good idea.
The rain has stopped so it's time to leave the forum and get on with my life...
In summary, if I'm after the last few percent of the faintest of signals then I'll leave the RBI flood off (and work more slowly) but not otherwise. The only drawback is that I need a full set of dark frames for the two scenarios and as I'm building a library for exposure lengths of 1500, 1000, 200, 40 and 8 seconds that's a lot of extra work as I'm aiming for 25 darks per setting. That's 250 darks and nearly twice that many discarded pre-exposures but fortunately MaximDL automates a lot of the image capture for me. At least with the Moon or clouds in the way I can feel I'm doing something