I missed most of this thread, so this might already have been covered.

When making a filter for CW, it's essential that you pay careful
attenuation to the time-domain response, and don't concentrate
exclusively on the frequency domain. Otherwise, you're apt to end up
with a filter that rings badly, and makes it difficult or impossible to
copy CW. If you're a very slow speed operator, you can put up with more
ringing, but at higher speeds, good transient response is vital.

Among the canonical filter types, I've found the Butterworth to be a
very good compromise for CW. It provides tolerable transient response
while having a respectable rolloff characteristic. In practice, a more
optimum design (i.e., equal sharpness and transient response with fewer
total poles) can often or always be realized with a combination of a
sharper filter type like a Chebyshev or even elliptical, with added
allpass poles to correct the phase response and therefore tame the
ringing. However, the latter approach can be a good deal more time
consuming -- you can look up the pole locations for a Butterworth in a
table, synthesize the filter, and have a reasonable expectation that
it'll work pretty much as designed. On the other hand, a phase-corrected
Chebyshev or elliptical filter can involve a lot of trial and error
modeling.

Roy Lewallen, W7EL