I'm dealing with tiny differences between small values and I'm using an inadequate instrument for the purpose, but I'm trying to get the best most reliable result possible under the circumstances. If this waveform or similar ones from this project can be shown consistently to be "negative" in area, even by a tiny bit, that would justify deploying the "big guns" for a further examination at greater and more reliable precision.
Instead of big guns get a small complementary gun of your own. You need something 8bit that has good floating point math (averaging!) or something with high bit count. Since frequencies are very low id say AD2 or PS 2204A which costs about as much as single good probe. They have nothing on Z for high freq work and Z has nothing on them where your interests lie.
Edit: I did look briefly what AD2 and my old PS 2205 offer here. PS does not have built in "area" functions, but does integrals and derivatives. AD2 has nothing like that built in, or I did not find it, but it is not a fundamental problem because you can create needed graph function with script straight in the GUI. 2205 has 16k memory while AD2 2x16k. Seems nothing, yet it is all used for calculus which results in somewhat different approach: Find balance between sampling rate and timebase that would fit max amount of wfms. Look for trends.
Example: 1.8Vpp triangle in from signal gen, slight DC offset set. Already simple DC Average measurement shows that there is more stuff above the zero, so does the trend of integral. Peak values taken from first cycle. With more powerful scope you would zoom out even further. Measurement values are much more accurate than 8bit or N pixels because untied from screen resolution and true floating point.
Of course to really get to bottom of stuff like this you would need to measure both voltage and current to account for reactive component.
You could try to zoom further out on Z also to see if there is any trend with integral but resolution would be quite low. Actually your 3 cycles already show a trend, but area measurement on the other hand do not so results are questionable, this is direct result of limited resolution.