The character from the unaffected part of the original value isn't random. It's actually very determinate, as you can see from the below example.

$s='foobarstuv';
foreach $n (3 .. 5)
{
   for $v (qw[ a ab abc abcd abcde abcdef ])
   {
      $d = substr( $c=$s, 1, $n, $v);
      $f = substr( $e=$s, 1, $n) = $v;
 
      printf "4-arg: '%13s' (%5s) ... 3-arg: '%13s' (%5s)\n", $c, $d, 
+$e, $f;
   }
   print $/;
}

--------------
4-arg: '     faarstuv' (  oob) ... 3-arg: '     faarstuv' (  aar)
4-arg: '    fabarstuv' (  oob) ... 3-arg: '    fabarstuv' (  aba)
4-arg: '   fabcarstuv' (  oob) ... 3-arg: '   fabcarstuv' (  abc)
4-arg: '  fabcdarstuv' (  oob) ... 3-arg: '  fabcdarstuv' (  abc)
4-arg: ' fabcdearstuv' (  oob) ... 3-arg: ' fabcdearstuv' (  abc)
4-arg: 'fabcdefarstuv' (  oob) ... 3-arg: 'fabcdefarstuv' (  abc)
 
4-arg: '      farstuv' ( ooba) ... 3-arg: '      farstuv' ( arst)
4-arg: '     fabrstuv' ( ooba) ... 3-arg: '     fabrstuv' ( abrs)
4-arg: '    fabcrstuv' ( ooba) ... 3-arg: '    fabcrstuv' ( abcr)
4-arg: '   fabcdrstuv' ( ooba) ... 3-arg: '   fabcdrstuv' ( abcd)
4-arg: '  fabcderstuv' ( ooba) ... 3-arg: '  fabcderstuv' ( abcd)
4-arg: ' fabcdefrstuv' ( ooba) ... 3-arg: ' fabcdefrstuv' ( abcd)
 
4-arg: '       fastuv' (oobar) ... 3-arg: '       fastuv' (astuv)
4-arg: '      fabstuv' (oobar) ... 3-arg: '      fabstuv' (abstu)
4-arg: '     fabcstuv' (oobar) ... 3-arg: '     fabcstuv' (abcst)
4-arg: '    fabcdstuv' (oobar) ... 3-arg: '    fabcdstuv' (abcds)
4-arg: '   fabcdestuv' (oobar) ... 3-arg: '   fabcdestuv' (abcde)
4-arg: '  fabcdefstuv' (oobar) ... 3-arg: '  fabcdefstuv' (abcde)
[download]

The extra characters are the characters from the string after substitution, up to the length of the substitution. So, the behavior is quite determinate. As well, it's very consistent. It's just a little counter-intuitive. *shrugs*