This is not really about hard versus soft references, but about different data types that happen to have the same name. In the statements
    $removed = [ @removed ];
and
    $removed = \@removed;
the scalar  $removed is the same thing: a hard reference to an array. In the first case, the reference is to an anonymous array created | initialized by a shallow copy from the  @removed array. In the second case, the reference is directly to the  @removed array. Again, both references are hard references, not soft (or symbolic) references. There are no soft references in the OPed code.

It is a "feature" of Perl that the variables  $foo @foo %foo &foo and maybe a few others are all different data types which can be made to have the same name (usually a practice to be avoided, IMHO).

Update: Here's some code that might give a better idea of what's going on:

c:\@Work\Perl\monks>perl -wMstrict -le
"use Data::Dump qw(pp);
 ;;
 my @ra = qw(a b c d);
 print 'A: ', pp \@ra;
 ;;
 my $hardref = [ @ra ];
 print 'B1: ', pp $hardref;
 pop @{$hardref};
 print 'B2: ', pp $hardref;
 print 'B3: ', pp \@ra;
 ;;
 $hardref = \@ra;
 pop @{$hardref};
 print 'C1: ', pp $hardref;
 print 'C2: ', pp \@ra;
"
A: ["a" .. "d"]
B1: ["a" .. "d"]
B2: ["a", "b", "c"]
B3: ["a" .. "d"]
C1: ["a", "b", "c"]
C2: ["a", "b", "c"]
[download]