If you have cultivar X, a diploid and the converted tetraploid version of that cultivar then comparing pollen sizes or guard cell sizes will identify which is diploid and which is tetraploid. A tetraploid cell is approximately twice the volume of the diploid cell which means that any measurement (length, width or height) of the tetraploid cell is about 35% longer than the diploid cell of the same cultivar.
Unfortunately cell size varies between different diploids and it varies between different tetraploids. The result is that some diploids have cells that are larger than some tetraploids (and vice versa some tetraploids have cells that are smaller than some diploids). The two distributions overlap.
The only observations that I know of on cell sizes in different ploidy daylilies did not look at a random sample of cultivars (they only looked at what they grew) and missed looking at both small-flowered diploids and tetraploids and longish-petaled diploids and tetraploids. I have been told that spidery daylilies have cell sizes that are 'different' (I assume from the diploids that were measured). So one can place little reliance on how much the measured cell sizes overlapped. I suspect that cell sizes are not only affected by ploidy but also by other factors - perhaps a general measure of plant size/robustness (perhaps for guard cells by leaf thickness and for pollen by flower, petal, anther thickness, etc).
What is probably more important is that triploid cell sizes will be intermediate between diploids and tetraploids and unfortunately completely overlap with both. A small cell size then might identify a plant as a diploid or triploid (but not perfectly from a tetraploid) while a large cell size might identify a plant as a triploid or tetraploid (but not perfectly from a diploid).
To get to a point where one could identify some cell sizes as diploid or triploid (not being able to identify which ploidy and with some of those being tetraploid - that is, as an example, after the cells of cultivar Y were measured it was possible to say that cultivar Y is probably a diploid or triploid but might be a tetraploid) and other cell sizes as triploid or tetraploid (without being able to identify which ploidy and with some of those being diploid, as an example, after the cells of cultivar Z were measured it probably is a triploid or tetraploid but might be a diploid ) one would probably need to measure the cells of several hundred diploid cultivars and several hundred tetraploid cultivars to build a good frequency distribution of cell sizes.
It may be possible (no guarantee) that triploids have a wider range of cell sizes than diploids or tetraploids. Statistically that would be described as triploids might have a larger variance in cell size than diploids or tetraploids. That is a statistical measure requiring statistical analysis and large numbers of cell measurements of known diploids, triploids and tetraploids to lay the foundation. To identify a cultivar as triploid would then require measurement of a large number of cells and statistical analyses comparing its variance with that of the known diploids and tetraploids. If in analyzing the known diploids and tetraploids one found that their size variances overlapped then one would be no better off than with the original measurement (the average or mean value).
