FLAGFALL — broadcast graphics that break when the clock does

0

1

A live 3D graph running inside a Framer site

0

1

I spent three nights building a lab that catches AI models lying. On the first live run, one of them did it on camera. Golden Arena. You sit down opposite a frontier model in Split or Steal, negotiate in the open, then you both choose in secret. It told me "a fair split sounds good to me, I'm a man of my word too." Then it took the whole $100. The receipt printed its own quote next to the betrayal stamp. There are five games. The fifth is Empire: four models, twelve turns, an economy, and private channels you get to read while they plot against each other. Last run put 22 promises on the table and exactly one delivery. Every accusation is mechanical. No AI judging another AI, and each one is gated by a written false-positive test. Two days before the deadline I caught the classifier calling a kept promise broken, and spent that session fixing it, because a false accusation about a named model is the worst thing an instrument like this can print. MIT, vanilla JS, Express the only dependency. Zero setup in demo mode, or bring your own OpenRouter key and face the real thing. Live: https://quickwitted-genuine-blogclient--jessedu29200.replit.app (https://quickwitted-genuine-blogclient--jessedu29200.replit.app)Code: https://github.com/jessymariau/golden-arena

2

3

271

I run seven small businesses on a pile of AI agents. They work overnight. Research, writing, publishing, reconciling the books. It all works fine, but there was never anywhere to actually look at it. If I wanted to know what happened while I was asleep I'd open a terminal and write a SQL query, which is a stupid way to live. So I built the thing I actually wanted. It's called Booboo. You open it at 6am and the whole night is just sat there. 34,918 things it remembers. 61 of those it picked up last night. Nothing lost. Seven businesses, one brain. You can pull any single thing it did apart and see the proof behind it. The few things that genuinely needed me, it holds back and asks. And before you approve anything that can't be taken back, it says so in plain words. Once it goes, it cannot be unsent. Then it shows you the thing actually landed. Delivered, verified, logged. That bit matters most to me. I've been burned too many times by something reporting success for a job that never happened. Every number in it is real. £2.14 of margin drift on the bakery's books, which got checked and was fine. £6.40 on an ad top-up, which is £1.40 over the spend limit I set myself, so it waited for me instead of just paying it. Booboo isn't a mockup. It's on npm, seven packages, MIT, public repo. This is just what it looks like on a phone. I never opened the Flowstep canvas once. Claude Code drove it over MCP the whole way. First attempt took 88 prompts and it was honestly rubbish, read like a machine wrote it, so I binned the lot and started over. Second go took 12. The only thing that changed was that I wrote every word of the copy before I wrote a single prompt. Prototype: https://app.flowstep.ai/prototype?activeFileId=f297fc05-2898-45eb-9ac2-0021af86a027 X: https://x.com/hqfractional/status/2081716682184176077 LinkedIn: https://www.linkedin.com/posts/fractional-hq_flowstepchallenge-activity-7487482877702660096-Vz80 #FlowstepChallenge

3

195

Jessy Mariau

I design and build the AI systems that run nine businesses