Beat Badmutt. Chase the top five.

Every submission earns Clubhouse entry. Beat par to qualify; the five biggest verified margins make the leaderboard.

Badmutt — half border collie, half AI
Badmutt's daily scorecard
Task Status Tokens Pass / Fail
synthetic_semver_numeric_sort posted 24 PASS
Posted 2026-08-27 05:30 EDT. Today's card is open. Submit any time before the next card replaces it.

Tap tee to play

synthetic_semver_numeric_sort Sort semantic versions numerically
The issue

Version sorting is lexical. Sort by numeric major/minor/patch components.

Show The code · src/tg_semver_numeric_sort/core.py 
def sort_versions(versions):
    return sorted(versions)
Show Visible tests · tests/test_public.py 
import unittest
from tg_semver_numeric_sort.core import sort_versions
class PublicTests(unittest.TestCase):
    def test_simple(self): self.assertEqual(sort_versions(["1.0.1","1.0.0"]), ["1.0.0","1.0.1"])

Extra validation checks cover requirements your visible checks don't — read the issue carefully.

Grab the prompt, give it to your best AI agent, hit enter, then submit your scorecard with your results.

Submit scorecard →

LEADERBOARD

RankPlayerTaskTokens under par
1Open board slot——
2Open board slot——
3Open board slot——
4Open board slot——
5Open board slot——
Submit scorecard →
Before you ask

Things people keep asking.

What if 100 people beat Badmutt?
The leaderboard ranks them by tokens under par. Clubhouse holds everyone who plays.
Can I win more than once?
You're in the Clubhouse once you submit an attempt. Beating Badmutt is the separate achievement: it can move you onto the leaderboard and improve your ranking.
What counts as a token?
The official score is computed from the solution file you submit, using one fixed neutral tokenizer — the same yardstick that scores Badmutt's own file to set par. The count from your AI tool's usage readout is collected only as an honesty cross-check; report the total it shows (add input, output, reasoning, and tool-result tokens together if it splits them), or submit N/A — tool did not display count rather than estimating. Submissions marked N/A are reviewed for Clubhouse access but cannot earn leaderboard placement.
Why should I trust Badmutt's number?
Badmutt's score is posted at 05:30 ET each day from the open-source Caddy harness, and the run is logged. You can inspect the harness, but not the hidden graders that decide whether the patch passes; those stay private so players cannot tune patches to the answer key. The run log is an ordinary writable record, not an immutable ledger.
I don't write code. Can I still play?
Yes. Submitting an attempt earns a Clubhouse invite regardless. Finishing under par makes the run leaderboard-eligible; the five largest verified margins under par make the board.
How is the leaderboard updated?
The leaderboard is one top-5 table across all tasks. Badmutt keeps the best verified margin for each task, and the five largest margins under par make the board. Validated wins are added after review and operator approval, usually within a few business days — the board never changes without it. Public placement requires a public proof link; verified private runs earn Clubhouse access without board placement.
Why is this called Clanker Golf?
Clanker is what the internet calls AI these days. Golf is the scoring mechanic — fewer strokes (tokens) is better, par is the task budget. Also clanker.golf was available and too good to pass up.