* Load an audio file (preferably one with some structured frequency content, e.g. with a strong tone at a known freq). In my case it was birdsong recordings, WAV, 44.1 kHz.
* Get a spectrum (not a spectrogram). Identify a notable feature of the spectrum at a short-ish resolution (in my case it was a dropoff due to a shelving filter at around 18 kHz, easily visible in an FFT of size 8192). I do this by moving my mouse over the feature, and looking at the visual readout of bin number and frequency.
* Change window size to 262144.
* Mouseover the feature again. The frequency is not reported correctly. For me the bin number was 111485--111616, which seemed about right, but the frequency was reported as 2370--2392 Hz (it should have been something around 18 kHz).
I can manually calculate what the real frequencies should have been reported as (assuming the bin indices are OK): 18754--18777 Hz. So about a factor of 8 wrong.