Turing.gun: human or AI?

Live: https://theguysaccount.github.io/turing-gun-challenge/ · Screens: docs/screens

Ten rounds. Each shows a question and two answers: one written by a person, one by an AI told to sound like a person. Pick the AI. After each round you see which was which and what gave it away. Scores share as a challenge link that replays the same game.

• Real human answers, quoted verbatim from databricks-dolly-15k (CC BY-SA 3.0), whose writers were told not to use AI. Each round cites its dataset row.
• AI answers written by Claude Opus 5.5 to the same questions, at similar length.
• Weekly game: everyone gets the same ten rounds each week, so scores compare.
• No server, no tracking. Guesses stay in the browser.

How it runs

What	Where	When
Tests, build, three headless-Chrome games, deploy, live games, IndexNow	.github/workflows/pages.yml	every push
Live games, page and data checks, keep-alive	.github/workflows/selftest.yml	weekly
Re-verify human answers against Dolly, license and links	.github/workflows/sources.yml	monthly

Details, fairness rules and the keys (new AI rounds, prize round, sponsor, domain, analytics): OPERATIONS.md.

Run, test, deploy

python3 pipeline/build.py && python3 -m http.server -d dist 8000
python3 -m unittest discover -s tests
node --test tests/js/*.test.mjs
scripts/e2e.sh
python3 scripts/check_sources.py

Deploy is a push to main.

License

Code: MIT. Round data (data/rounds.json): CC BY-SA 3.0, because the human answers come from databricks-dolly-15k (CC BY-SA 3.0, © Databricks).