If you sample a nominally 100 MHz signal at precisely twice its frequency, as you might if you derive your sample clock from the input signal, say, with a PLL that doubles the input frequency, or a using a differentiated version of the input signal, thereby sampling at each point at which dv/dt changes in sign, you could, indeed wind up with a set of samples that never reach the minimum or maximum of the amplitude. If, however, you sample at a frequency that is not precisely coherent with the input signal, there will be a phase shift due to the lack of synchronization between the input signal and the sample clock, and that will, given enough samples yield a decent approximation of the input signal's amplitude. Only with some detailed foreknowledge of the nature of the input signal can you go any further. If, for example, you know it to be of a specific type and symmetry, you can, in your firmware, draw conclusions not directly observed.

For some time, sampling oscilloscopes gathered thousands of samples at a sample rate known to be incoherent with the sampled signal just to produce a good display of the waveform. In many respects, this is exactly what the analog oscilloscopes did, though they simply displayed repetitions of the input. The result was always a satisfactory representation of the waveform under observation. You can always count the input, synthesize something close but NOT exactly the same, and then proceed.

RE