Thank you for your answers.

OK so AGC must be off. What made me confused is the fact that the
manufacturer explicitly indicates that this alignment must be done
with AGC auto.
With AGC off, the gain can be increased up to a point where the curve
gets clipped and wider, this is the point then. A scan width of 1sec
seems to be the limit for an optimal representation.

The receiver has a proper IF output, it is just before the AM detector
diode but not immediately after the filter, in between there is a
three stages, fixed bandwidth, high gain IF amplifier.

With AGC off the output level now varies depending on which IF filter
is selected: the manual says that the input level should be regulated
for 0dB at IF output, but this is assuming AGC on. Although I always
try for maximum gain, each filter seems to produce its own maximum
output level, and differences are as large as 15dB+ with no obvious
logic, so no the wider the filter the higher the level. Does this make
sense? How can I detect which is the "reference" level to look for or,
alternatively, what is an "acceptable" difference among the filters?

Thank you and best regards
Paolo