Details about algorithm and "10 cycle" rule of thumb
Reference is the online manual: https://manual.dewesoft.com/x/setupmodule/modules/general/math/freqdomainanalysis/exactfrequency
Answer
Exact frequency uses an FFT-based approach, and it splits the interval into two intervals.
So when the user selects something equivalent to 10 periods, the dF is based on just one half of that interval (dF is 20% rather than 10% of the frequency of interest).
This could be improved in the sense that we could use a 75% or 90% overlap of the two windows where we calculate the frequency.
10 periods is maybe "a bit too much", but going below 4 periods gets really useless really soon. Both in the sense that accuracy gets large, but one can also get very weird side-effects from other nearby frequencies, or even "frequency jumps" when the calculated frequency is "completely off" (there's some modulo 2*pi ambiguity, and with a small number of cycles and zero overlap one can easily use the "wrong modulo").
Providing the exact limit is not easy because this of course depends on whether one has a purely sinusoidal signal, or a lot of noise (it behaves worse in noisy conditions).
But generally (somewhat experimentally): if one half of the interval contains less than 2 periods (which is 4 periods for how the exact frequency is currently implemented), the results won't look good. If it contains less than 1 period, it's totally totally useless. With around 4 periods (so 8 periods for exact frequency) the results get really nice / precise / stable. (By modifying exact frequency math, we could improve the functionality at shorter intervals a little bit.)