"New normalize filter"?

Recently switched from old Audacity 2 to current 3. Today I made an outdoor recording that was too silent (but alas 24bit!), so I tried to “boost it a bit” using normalize:
Playing the clip, the peak level displayed was something like “-24dB” (RMS analysis says something like -40.5dB), so I tried to normalize to “-15dB”.
Surprisingly the clip was even more silent than before, and the waveform preview was “shrunk”. So I wonder whether something has changed, or whether this is a bug.
Actually I had imported 5 WAV tracks, and “solo’d” one, marked it completely, and then applied the normalize filter. Could the existence of multiple tracks confuse the filter logic?

Audcaity3 has Loudness Normalization which uses RMS or LUFS, not peak.
If you want consistent loudness RMS/LUFS are more reliable than using peak values.

I understand that Loudness Normalization works differently, and when I use that to normalize to -20dB, the peak is somewhere near -12dB (it becomes significantly louder). However originally I used the old(?) plain “Normalize” that should work the same as in version 2.4.2, right? Somehow I feel it’s broken now.

“plain Normalize” uses peak values.
If there is one loud peak (e.g. at the start or end where you don’t notice),
loud peak near end
That one peak determines the loudness of the track after plain normalize.

Seems you are right: Probably a classic “left to right reading problem”:
You start looking at the left, listen to the start and realize that it’s not loud enough.
The “plop” at the end is most likely when I had pressed the button to stop the recording…