use utf8;
use open ':std', ':utf8';
use feature 'unicode_strings';

...

for my $file ( @files ) {

    open my $FH, '<', $file or die "Cannot open '$file' because: $!";
    my $size = read $FH, my $data, -s $FH or die "Cannot read from '$f
+ile' because: $!";
    $size == length( $data ) or die "Error reading from '$file'\n";
    close $FH;

...
[download]

This will always result (if $data contains Unicode characters) in dieing with the message "Error reading from '$file'\n".

Unlike in OP, here with perl v5.36.3/FreeBSD 14 (with the code below) the numbers from length & read functions match (226); do not obviously match the size in byte (282). Is it matter of the perl version; and/or, is it how one is holding the encoding layer?

perl check-size.pl ./data.mixed
$VAR1 = {
          '-s' => 282,
          'length' => 226,
          'read' => 226,
          'tr' => 226
        };
[download]

# Usage: perl5.36.3 check-size.pl ./data-mixed

use v5.32;
use warnings;
use warnings qw[ FATAL utf8 ];
use open qw[ :encoding(UTF-8) :std ];

use Data::Dumper;
local $Data::Dumper::Sortkeys = 1;

my $data = $ARGV[0]
   or die qq[Give a file to count sizes of different kinds.\n];

my %stat = ( q/-s/ => -s $data );

open( my $fh, q/</, $data )
   or die qq/Cannot open $data to read: $!/;

my $string;
$stat{q/read/} = read( $fh, $string, $stat{q/-s/} )
                  or die qq/read from \$fh failed: $!/;

close( $fh )
   or die qq/Could not close $data: $!/;

$stat{q/length/} = length( $string );
$stat{q/tr/} = $string =~ tr///c;

print( Dumper( \%stat ) );
[download]

Base64-encoded generated data -- if anyone would rather have it than run above "make-data.pl" -- with SHA256 sum of a1e81919b72403bd3c9d95979dae8d928354cf64de8d999bcad00aec4298cf76 (decoded value has the sum of 02bec20353169b58ea51558fa7d6c4316bd13f5e4a3d976484e8ed7700935962) ...

Tj9EOzg3aHosYU9NbDtrLmZQcUt4M3l3dl8hR1IyLEpvNHJCMVdRVTVUaVpkQ2pMSVM2Xz
+lFY1hl
dUhuIXRtLlkwRlZzZ3A/QS0tYk4/RDs4N2h6LGFPTWw7ay5mUHFLeDN5dwpGNVRuRUtsOG
+rwn5it
MGhQTzHvuI/ig6Mxc0lvQUx24KWsUyAgcHo2M0NiWuClr/CfpK9XVjJrR23wn5iSUnU3VW
+NI4KWv
YXFneDR34KWnWPCfmI85ZDHvuI/ig6Pwn6SjZk5Z8J+YrXlNSuClpmXgpajgpaZp4KWqdF
+FyREJG
NVRuRUtsOGrwn5itMGhQTzHvuI/ig6Mxc0lvQUx24KWsUyAgcHo2M0NiWuClr/CfpK9XVj
+IK
[download]

Update: *ugh*...

• The Base64-encoded sample has different checksum after downloaded from here (than on the machine where it was posted from). The checksum for the decoded value is still the same as posted.
• Removed the program that produced the sample Unicode output as I had apparently changed it (during the (re)editing of this post) such that it produced different output than the above mentioned Bas64-encoded text (for which the size result is shown).