Oh yeah, that can happen when you heavily clip. You seem to have droven your output so high, that there a several samples with 0dbfs in a row. You have to see it from the point of a DAC. after the samples there is a reconstruction filter that gets rid of the 'steps'. Imagine it as smooting out of the steps from the bits, creating one continuus waveform.
Image 4 samples as bars, first -60dbfs then 2 samples 0 dbfs and the fourth one -60dbfs again. No draw a line and make a smooth waveform over those 4 samples. See? You will overshoot at those two 0dbfs samples to get to a smooth waveform.
True peak means "all right, we do not measure what any of those samples value is. we measure what the actual waveform would look like after it is converted back to analog." this is not at all about different compression algos that is just what needs to be done to get the true (after DAC) peaks of any signal that came from digital and went to analog. But lossy compresson can make it worse, or fuck it up in other ways.
Unfortunately I can't tell you what spotify does when you upload such a track.
Could you find out by uploading it and recording it back?
Would be highly interesting.

//edit// oh my dark lord, I replied to an ancient post. never mind

 

Lol, it's ok. I'm actually glad you guys resurrected this topic

So I tried Loudness Penalty site and plugin and it's not very helpful to be honest. I see that it just takes integrated LUFS of my wav file and calculates the difference between -14 LUFS and my value. So if I have -6, it shows -8 spotify penalty. I can do that math myself.
I hoped the plugin would do it this way:

Meaning they would do ogg or whatever conversion, emulating each service's real pipeline. Because two different tracks with the same -0.1 dbtp will most likely have different penalties after their conversion

Anyway, I remembered that last time I uploaded my track I had to reuipload it with -1dbtp setting instead of 0dbtp. It helped, but not much. Might actually try -2dbtp next time