[hannes69]

I've been staring at my plasma in disbelief for ten minutes...

Wonders can happen

ELEVEN years I've been HTPCing and I understand this now...

It is not too difficult to understand when you think about the whole chain from beginning to end. In my opinion it is often described only half the way or too scientifically in the internet. Practically it is some piece of logic.
Better after eleven years than never

Clock deviation has now gone to 0.00242% (positive).

I have observed that the deviation is temperature dependant (no joke!). But no wonder, it´s all electronics and so physics and so often temperature dependant. The value of the deviation in madvr stabilizes very fast in time (I´d say within 5 minutes). Now compare the value of the deviation with a cold fresh boot computer versus a several hours warm run computer (with graphic card used by e.g. watching a movie). You will detect a deviation of the value of the deviation So take a mean value to be absolutely accurate (or take one custom resolution for summer and one for winter time )

If I multiply 23.9760239 for 1.0000242 I get 23.976604196. My refresh is 23.97656, close enough to get these results (1 frame drop every 4.41 hours).

Is the reasoning correct? So frame drops for positive click deviation, frame repeat for negative clock deviation?

Your measured refresh rate is 23.97656Hz. Your ideal refresh rate is 23.97660Hz (within the number of decimates in madvr). So your measured rate is lower than the desired ideal rate. So in reverse your refresh period = 1 / measured refresh rate is too high = too long. These very little too long frames accumulate over time and in your case after 4.41 hours exactly one frame accumulates which now has to be dropped.
Short: measured refresh rate < (movie framerate * (1 + clock deviation)) -> frames drop
measured refresh rate > (movie framerate * (1 + clock deviation)) -> frames repeat

Warner you cannot act on the quartz themselves. This basically "cheats" your video card into providing the correct result by asking for a refresh rate that compensates for the two clocks difference.

Nice statement, this shows that you have understood the topic

BTW I just discovered that for 50Hz my GTX660 is outputting 50.00123 by default which pretty much compensates the clock deviation to perfection (no frame-drops/repeats expected).

Sometimes one has pure luck No work for this refresh rate...

But I seem to be unable to create a custom 59.940 resolution through the control panel (whereas I had no problem creating the 23.976 one). Nvidia *still* seems unable to implement this portion correctly after so many years

Depending on your output device you can of course try n * 29.97Hz with n >= 1
Or you can use CRU utility which does the same and you don´t need to use the graphics vendor stuff.

One more hint:
You can use THIS PixelClockCalculator (highly recommended) to calculate exact values. If you are working with too many decimal places (e.g. 6) and get no results, try to lower it to 5 or 4. the "Parameters" depend on your device, for a first try leave them as they are.
Then, maybe you may detect, that the calculated refresh rate of this tool and when using it the shown value in madvr differ. This is due to the fact, that AMD in my case has problems to implement this thing right. I don´t know how accurate the results are with nvidia and intel. With my AMD card I get constant offsets of the calculated numbers, so I´m simply adding this offset at calculation time of the tool. You have to find out by yourself if you have such an offset.
If you spend some time on this and have some luck with your output device which should "eat" (how we in Germany say ) all custom refresh rates, you will get "1 frame drops/repeat every year" in madvr (actually it shows ... days).
Good luck!

Bytheway ashlar, what´s your output device that works with the *1 multiple, it´s rare that a device works with refresh rates < 30Hz... Is it a projector?

so the damage warning is just for the lolz?

There are people who like to be on the SAFEST side. "DO NOT DRY YOUR CAT OR DOG IN THE MICROWAVE OVEN"
I am an electric engineer and know some of the electronics stuff. So: Your output device can do some crazy optical things or freeze/hang, but after repowering (power off - power on) everything is fine again.
Bytheway overclocking of computer components by itself doesn´t do the harm (I speak of processor, GPU, memory). If you clock too high it errors. Nothing more. The danger comes from too high voltage or too much heat. If you simple rise the clock you very soon come to a point where you have instability. So people rise voltage so rises the heat so the components are grilled. The overCLOCKING part is totally secure (if temperature is within allowed limits).
And thats the same logic with our custom resolutions. No voltage, no heat. And in this case even no OVERclocking, but some kind of exacter clocking for our wishes.
A damage warning in this context exists for the people who expect such a warning placed