This appears to be a bug in the function coeff, called from resm1 in src/residu.lisp. AFAICT, the taylor series returns the correct value, but coeff is unable to get the coefficient of 1/(x-p).

A workaround is the remove the call ($ratsimp ($rectform pole)). This fixes both this bug and bug 3541292, without causing a regression in bug 1504505.

But coeff should really be fixed to return the correct value.