Formula Challenge

Patrick2788 even though this isn't one of your challenges it sort of is.  I don't have those fancy new functions so I just used a SCAN and FILTER with a twist:

=LET(peaks,SCAN(0,VSTACK(torque,0),LAMBDA(p,q,IF(q<=0,IF(p>=25,-p,0),IF(q>p,q,p)))),
     FILTER(-peaks,peaks<0))

The inner SCAN does a basic max function by choosing the new value if it is > the old value with a slight twist being if the new value is <=0 and the previous value was a peak (>25) then it assigns the output as -p (negative of the held max).

The filter then just filters for those 'negative max' values and negates them

My results are a bit different than yours because there are cases where the values jump negative and then back positive.  Possibly a noise filter on the data is needed to filter that out.  

Alternatively if you replace the IF(q<=0 part with IF(q=0 then my results are nearly identical with yours with the only exception being your output has 1 more peak at the end that my formula (and my eyes) doesn't find.  But again this version works because it relies on those 0s and if the data was more sinusoidal and could 'jump' over the 0 crossings to negative values the above formula with <=0 would be more appropriate.