Submitted for the June Solstice Game Jam — Best Ode to Alan Turing category
I got flagged by Sloan.
If you've been on DEV long enough, you know Sloan. I thought Sloan was a bot. Sloan is Francis — someone I've exchanged comments with for months, since before Richard left the platform. He posted about the flagging openly, tagged the founders, explained his reasoning. Then added: "This was hard to tell you for many reasons." He reads every flagged article himself, runs it through GPTZero, makes a call. He knew me. He flagged me anyway.
One of the flagged articles had sparked a five-exchange comment thread that became an open-source repo. The thinking was mine. The flag still landed.
That's the uncomfortable thing about the Turing Test in 2026: it doesn't measure origin. It measures surface texture. And if you write well enough, you sound like a machine.
The Game
Five questions. You write your answers. Claude scores them 0–100 on how human they sound, tells you what gave you away, and at the end gives you an average.
The questions are the ones that actually separate humans from pattern-matchers:
- Describe the last time something genuinely surprised you. Not shocked — surprised.
- What's something you changed your mind about in the last year? What moved you?
- What's a skill you have that you never bothered to put on your CV?
- Name something you've read or watched that you think about more than you expected to.
- What do you actually think about AI? Not what you're supposed to think — what you actually think.
The scoring prompt at the heart of it:
Human signals: named specifics, opinions that could get you in trouble, genuine uncertainty, things slightly off-topic but revealing.
AI signals: balanced framing, hedge words, smooth transitions, excessive completeness.
Score 60 or above: Passes. Below 60: Flagged.
The Solstice Angle
The June solstice is the longest day — the day the sun is most itself. Unambiguous. No hedging.
That's what this game is asking for. Not your best answer. Your most you answer. The one that's slightly off-topic, slightly embarrassing, slightly specific in a way that reveals something you didn't mean to reveal. That's the signal.
Turing's original question was: can a machine think? The question we're living with now is its inversion — can a human still sound like one?
Technical Approach
Vanilla JS, no dependencies, one HTML file. Cloudflare Worker as proxy, Pages for hosting.
The Worker sits between the browser and Anthropic. It receives your prompt and response, calls the API, returns { score, verdict, reason }. The frontend never sees the API key. One call per question, nothing stored, model is claude-sonnet-4-6.
The frontend is a single index.html — progress bar, animated score fill, final breakdown screen. No build step. A judge can open DevTools and follow exactly what happens on each submit.
proof-of-human/
index.html ← full game
worker/
src/index.js ← CF Worker proxy
wrangler.toml
What I Learned Building It
The scoring prompt took the most iteration. The first version was too generous — everything passed. The second was too harsh — everything got flagged. The final version keys on one thing: specificity that costs you something. An answer that names a real person, admits a real mistake, or takes a position you might regret. That's what the model now reliably catches.
The irony: I had to write like an AI to build a detector for AI writing. I kept second-guessing my own prompt phrasing, smoothing transitions, hedging. The game caught me too.
Play It
→ proof-of-human-3ts.pages.dev
Source: github.com/dannwaneri/proof-of-human
Built June 2026. Vanilla JS. One API call. No frameworks. The Sloan incident was real.
Top comments (3)
Hey @francistrdev . you're the origin story for this one. Built it for the June Solstice Game Jam after the flagging incident. Five questions, Claude scores how human you sound.
Curious what you'd score. → proof-of-human-3ts.pages.dev/
Hey Daniel!
Pretty much all are one sentence answers lol
72 . you passed. The 82 on Q5 makes sense, that's the question where one-sentence answers still carry weight because there's no safe answer. The 62s on Q1 and Q3 are the specificity gap . one sentence doesn't leave room for the detail that gives you away as human. Play again and go longer on those two.