^g-force
about the luma variations, sounds like brightness AGC. If so, deflicker won't touch it.
Fizick's DeFlicker page claims that it can fix AGC blinding if you set the Lag to a negative value. Also, I've verified that these brightness variations are on the source tape.
about the duplicate frames, IF 4 of the five captures always have the correct frame at any one time, the median-of-five script will take care of it for you automatically. The method you describe sounds like a LOT of work (you will get false positives that you will have to weed through, or you will miss some due to noise and other fluctuations)
There's no way I can be sure without watching all five captures very closely. I'd like to see how many duplicates MultiDecimate identifies before deciding on any single method.
about the left-hand curve, does it change based on overall brightness (i.e. change on scene changes)? Just thinking of power supply issues old CRTs had, maybe something similar in the capture equipment.
The lefthand-side curve seems unrelated to the brightness; its size and shape is random and non-periodic. For example, it'll be one size and shape for 5 minutes here, another for 7 minutes there, etc.
If you post some raw video that exemplify the problems you are describing, SOMEONE here will have an answer for you.
I will. I passed-out last night, so I didn't get a chance to.