diff --git a/contrib/watch-sched-debug b/contrib/watch-sched-debug
deleted file mode 100755
index edf2b2e..0000000
--- a/contrib/watch-sched-debug
+++ /dev/null
@@ -1,199 +0,0 @@
-#!/usr/bin/perl
-# -*- mode: perl; indent-tabs-mode: nil; perl-indent-level: 4 -*-
-# vim: autoindent tabstop=4 shiftwidth=4 expandtab softtabstop=4 filetype=perl
-
-use strict;
-use warnings;
-
-use Time::HiRes qw(time sleep);
-use Getopt::Long;
-use Text::Diff;
-use threads qw(yield);
-use Thread::Semaphore;
-use Thread::Queue;
-
-use Data::Dumper;
-$Data::Dumper::Sortkeys = 1;
-$Data::Dumper::Pair = ' : ';
-$Data::Dumper::Useqq = 1;
-$Data::Dumper::Indent = 1;
-
-my %args;
-$args{'interval'} = 10;
-$args{'cpus'} = { '0' => 1 };
-$args{'count'} = -1;
-
-sub arg_handler {
-    my ($opt_name, $opt_value) = @_;
-
-    if ($opt_name eq 'count') {
-        $args{'count'} = $opt_value;
-    } elsif ($opt_name eq 'interval') {
-        $args{'interval'} = $opt_value;
-    } elsif ($opt_name eq 'cpus') {
-        $args{'cpus'} = {};
-
-        foreach my $element (split(/,/, $opt_value)) {
-            if ($element =~ /^[0-9]+$/) {
-                $args{'cpus'}{$element} = 1;
-            } elsif ($element =~ /-/) {
-                my @array = split(/-/, $element);
-
-                for (my $i=$array[0]; $i<=$array[1]; $i++) {
-                    $args{'cpus'}{$i} = 1;
-                }
-            }
-        }
-    }
-}
-
-sub read_file {
-    my ($file_name) = @_;
-
-    my $timestamp = time();
-    my $file_content = do {
-        local(@ARGV, $/) = $file_name;
-        <>
-    };
-
-    return ($timestamp . '|' . $file_content);
-}
-
-GetOptions("interval=s" => \&arg_handler,
-           "count=s" => \&arg_handler,
-           "cpus=s" => \&arg_handler)
-    or die("Error in command line arguments");
-
-my $quit : shared = 0;
-my $data_queue = new Thread::Queue();
-my $signal_quit = 0;
-use vars qw ($signal_quit);
-
-$SIG{'INT'} = sub { $signal_quit = 1; };
-
-my $collector_thread = threads->create('collector', $data_queue, \$quit);
-my $reporter_thread = threads->create('reporter', $data_queue, \$quit);
-
-while(($signal_quit == 0) && $collector_thread->is_running() && $reporter_thread->is_running()) {
-    sleep($args{'interval'} / 2);
-}
-$quit = 1;
-
-$collector_thread->join();
-$reporter_thread->join();
-
-exit(0);
-
-sub reporter {
-    my ($queue, $quit) = @_;
-
-    my $cur_sample;
-    my $prev_sample;
-
-    while(!$$quit || $queue->pending()) {
-        undef($cur_sample);
-
-        my $cur_sample = $queue->dequeue_timed($args{'interval'} / 2);
-
-        if (defined($cur_sample)) {
-            $cur_sample = parse_sample($cur_sample);
-
-            if (defined($prev_sample)) {
-                diff_samples($prev_sample, $cur_sample);
-            }
-
-            $prev_sample = $cur_sample;
-        }
-    }
-
-    return(0);
-}
-
-sub collector {
-    my ($queue, $quit) = @_;
-
-    my $counter = 0;
-
-    while(!$$quit) {
-        my $sample = read_file('/proc/sched_debug');
-        $queue->enqueue($sample);
-
-        if ($args{'count'} != -1) {
-            $counter += 1;
-            if ($counter == ($args{'count'} + 1)) {
-                $$quit = 1;
-            }
-        }
-
-        if (!$$quit) {
-            sleep($args{'interval'});
-        }
-    }
-
-    return(0);
-}
-
-sub parse_sample {
-    my ($sample) = @_;
-
-    my %sample_data;
-
-    my @data = split(/\|/, $sample, 2);
-    $sample_data{'timestamp'} = $data[0];
-
-    my @cpus = split(/cpu#/, $data[1]);
-
-    for (my $i=1; $i<scalar(@cpus); $i++) {
-        $cpus[$i] =~ m/^([0-9]+),/;
-        my $cpu_num = $1;
-
-        if (! exists($args{'cpus'}{$cpu_num})) {
-            next;
-        }
-
-        my @cpu_data = split(/runnable tasks:/, $cpus[$i]);
-
-        $sample_data{$cpu_num} = $cpu_data[1];
-    }
-
-    return \%sample_data;
-}
-
-sub format_timestamp {
-    my ($ts) = @_;
-
-    my @var = localtime(int($ts));
-    my $subsec = ($ts - int($ts))*1000;
-
-    return (sprintf("%04d-%02d-%02d %02d:%02d:%02d.%03d", $var[5]+1900, $var[4]+1, $var[3], $var[2], $var[1], $var[0], $subsec));
-}
-
-sub diff_samples {
-    my ($start, $stop) = @_;
-
-    print "\n";
-    printf "Start: %s (%s)\n", format_timestamp($start->{'timestamp'}), $start->{'timestamp'};
-    printf "End:   %s (%s)\n", format_timestamp($stop->{'timestamp'}), $stop->{'timestamp'};
-
-    foreach my $cpu (sort(keys %{$args{'cpus'}})) {
-        print "\nCPU: $cpu\n\n";
-
-        my @start_array = split(/\n/, $start->{$cpu}, 4);
-        my @stop_array = split(/\n/, $stop->{$cpu}, 4);
-
-        print $start_array[1] . "\n";
-        print $start_array[2] . "\n";
-
-        my $cpu_diff = diff(\$start_array[3], \$stop_array[3], { STYLE => "Unified", CONTEXT => 0 });
-
-        my @diff_lines = split(/\n/, $cpu_diff);
-
-        for (my $i=0; $i<scalar(@diff_lines); $i++) {
-            if ($diff_lines[$i] =~ /^@@/) {
-                next;
-            }
-
-            print $diff_lines[$i] . "\n";
-        }
-    }
-}
diff --git a/contrib/watch-sched-debug.py b/contrib/watch-sched-debug.py
new file mode 100755
index 0000000..426bbf2
--- /dev/null
+++ b/contrib/watch-sched-debug.py
@@ -0,0 +1,169 @@
+#!/usr/bin/env python3
+# -*- mode: python; indent-tabs-mode: nil; python-indent-level: 4 -*-
+# vim: autoindent tabstop=4 shiftwidth=4 expandtab softtabstop=4 filetype=python
+#
+# Monitor the kernel scheduler debug file for changes by periodically
+# reading it and displaying a unified diff of the per-CPU runnable task
+# lists between consecutive samples. Useful for identifying unexpected
+# task migration, pinning violations, or scheduling anomalies on
+# isolated CPUs during benchmark runs.
+
+import argparse
+import os
+import difflib
+import queue
+import signal
+import sys
+import threading
+import time
+from datetime import datetime
+
+
+def parse_cpus(value):
+    cpus = set()
+    for element in value.split(","):
+        if "-" in element:
+            parts = element.split("-", 1)
+            for i in range(int(parts[0]), int(parts[1]) + 1):
+                cpus.add(i)
+        else:
+            cpus.add(int(element))
+    return cpus
+
+
+def default_sched_debug_file():
+    for path in ["/sys/kernel/debug/sched/debug", "/proc/sched_debug"]:
+        if os.path.exists(path):
+            return path
+    return "/sys/kernel/debug/sched/debug"
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Watch /sys/kernel/debug/sched/debug for changes")
+    parser.add_argument("--interval", type=float, default=10, help="Sampling interval in seconds (default: 10)")
+    parser.add_argument("--count", type=int, default=-1, help="Number of samples to collect (default: infinite)")
+    parser.add_argument("--cpus", type=parse_cpus, default={0}, help="CPUs to monitor, e.g. 0,2,4-7 (default: 0)")
+    parser.add_argument("--file", default=default_sched_debug_file(), help="Path to sched debug file (default: auto-detect)")
+    return parser.parse_args()
+
+
+def read_file(file_name):
+    timestamp = time.time()
+    with open(file_name, "r") as f:
+        content = f.read()
+    return (timestamp, content)
+
+
+def parse_sample(timestamp, content, watched_cpus):
+    sample = {"timestamp": timestamp}
+    cpus = content.split("cpu#")
+
+    for i in range(1, len(cpus)):
+        match = cpus[i].split(",", 1)
+        cpu_num = int(match[0])
+
+        if cpu_num not in watched_cpus:
+            continue
+
+        parts = cpus[i].split("runnable tasks:", 1)
+        if len(parts) > 1:
+            sample[cpu_num] = parts[1]
+
+    return sample
+
+
+def format_timestamp(ts):
+    dt = datetime.fromtimestamp(ts)
+    return dt.strftime("%Y-%m-%d %H:%M:%S") + ".%03d" % (dt.microsecond / 1000)
+
+
+def diff_samples(prev, cur, watched_cpus):
+    print()
+    print("Start: %s (%s)" % (format_timestamp(prev["timestamp"]), prev["timestamp"]))
+    print("End:   %s (%s)" % (format_timestamp(cur["timestamp"]), cur["timestamp"]))
+
+    for cpu in sorted(watched_cpus):
+        if cpu not in prev or cpu not in cur:
+            continue
+
+        print("\nCPU: %d\n" % cpu)
+
+        start_lines = prev[cpu].split("\n", 3)
+        stop_lines = cur[cpu].split("\n", 3)
+
+        if len(start_lines) > 1:
+            print(start_lines[1])
+        if len(start_lines) > 2:
+            print(start_lines[2])
+
+        if len(start_lines) > 3 and len(stop_lines) > 3:
+            diff = difflib.unified_diff(
+                start_lines[3].splitlines(keepends=True),
+                stop_lines[3].splitlines(keepends=True),
+                n=0,
+            )
+            for line in diff:
+                if line.startswith("@@") or line.startswith("---") or line.startswith("+++"):
+                    continue
+                print(line, end="" if line.endswith("\n") else "\n")
+
+
+def collector(data_queue, quit_event, args):
+    counter = 0
+
+    while not quit_event.is_set():
+        timestamp, content = read_file(args.file)
+        data_queue.put((timestamp, content))
+
+        if args.count != -1:
+            counter += 1
+            if counter == args.count + 1:
+                quit_event.set()
+                return
+
+        quit_event.wait(args.interval)
+
+
+def reporter(data_queue, quit_event, args):
+    prev_sample = None
+
+    while not quit_event.is_set() or not data_queue.empty():
+        try:
+            timestamp, content = data_queue.get(timeout=args.interval / 2)
+        except queue.Empty:
+            continue
+
+        cur_sample = parse_sample(timestamp, content, args.cpus)
+
+        if prev_sample is not None:
+            diff_samples(prev_sample, cur_sample, args.cpus)
+
+        prev_sample = cur_sample
+
+
+def main():
+    args = parse_args()
+
+    quit_event = threading.Event()
+
+    signal.signal(signal.SIGINT, lambda *_: quit_event.set())
+
+    data_queue = queue.Queue()
+
+    collector_thread = threading.Thread(target=collector, args=(data_queue, quit_event, args), daemon=True)
+    reporter_thread = threading.Thread(target=reporter, args=(data_queue, quit_event, args), daemon=True)
+
+    collector_thread.start()
+    reporter_thread.start()
+
+    while not quit_event.is_set() and collector_thread.is_alive() and reporter_thread.is_alive():
+        time.sleep(args.interval / 2)
+
+    quit_event.set()
+
+    collector_thread.join()
+    reporter_thread.join()
+
+
+if __name__ == "__main__":
+    main()