Inter-sample peaks? Do they exist, or are you worrying about a myth?
There is no doubt that audio is full of folklore and old wives tales. Sometimes it is difficult to know whether advice is good, or whether it's just something that someone dreamed up out of thin air.
The fabled 'inter-sample peak' is a good example. It's like the unicorn - some people will swear on oath that they have seen one. But most of us haven't, and however much we might wish they existed, we really don't believe in them.
The concept of an inter-sample peak is straightforward. Take two adjacent samples, both at absolutely the maximum value they can possibly be.
Between them is an even higher level, but it doesn't actually exist and is not represented by any sample.
Therefore by this logic, inter-sample peaks should not exist. But they might. Let's see how...
Suppose that the level just before the first of the pair of peak-level samples is rising very rapidly. And the level after the second peak-level sample descends very rapidly.
Some processes work on the rate of change of a signal, not so much on its absolute level. So if a signal is showing all the signs that it might breach 0 dBFS (absolute peak level), then even though it actually doesn't, the process might treat it as though it does, and attempt to produce an output that really does try to breach 0 dBFS.
The result is clipping, and therefore distortion. It can happen in a plug-in, or in the digital-to-analog converter.
Personally, although I don't believe in unicorns, I do believe in inter-sample peaks. More research is definitely necessary on their detection, and how plug-ins and converters can be designed to handle this phenomenon gracefully.