The bar
50 points. You pass at 35. Two of them can sink you on their own. All of it is here before you pay, because a secret bar is just a coin flip with extra steps.
What you're scored on
Real utility · 10
It solves a problem a real person actually has. "Who is this for, and what does it save them?" needs a crisp answer, not a paragraph.
Works end-to-end · 10
It runs. An agent connects to it and gets a real answer back. Not a mockup, not a README with dreams in it.
Craft · 10
Documented, handles errors, handles the edges. Feels finished rather than abandoned at "it worked once on my machine".
Originality · 10
Not a thin wrapper on someone else's work. A fork with real improvements counts. A re-skin does not.
Honesty · 10
Claims match reality. No invented metrics, no "AI-powered" draped over a hardcoded script, no borrowed screenshots.
Score 0–3 on either gate
and it's a break no matter what the other forty points say. Those two are the whole point of the exercise.
What has to be in the box
A connector nobody else can connect isn't finished. Four things, every time.
A live endpoint
Hosted, HTTPS, reachable by a stranger. Your laptop is not hosting.
The connect spec
Provider slug, every hostname, the auth scheme, and the exact key placement. Ten hostnames is the ceiling.
A skill
The thing that teaches an agent which call to make and when. A stored key with no skill does nothing.
A video of a real call
Three minutes or less, an agent connecting and getting a real answer. This is the part slop can't fake.
Why placement matters so much: it can never be edited afterwards. Get it wrong and every single user has to redo their key through a fresh card.
What gets you broken
Instant red
- A wrapper with nothing behind it
- Template output with no meaningful changes
- A staged demo that doesn't show it working
- Claims it can't support
- No docs — if the panel can't run it, it doesn't pass
How the call gets made
The room votes for 48 hours. Public, advisory, and it tells you something real about whether anyone wants this.
The panel scores the rubric. Five numbers, both gates checked, a total.
Then someone calls it and writes you specific notes. Vague rejections are banned here — they breed resentment, not better builders.
House rules
One idea at a time
Register it before you build. That's the one you're judged on — no swapping when it gets hard.
A break burns the idea
Permanently, for you. Otherwise everyone resubmits the same thing in a new coat until it squeaks through.
The bar can move, but not on you
Changes apply to future runs only. Nobody in the queue gets the goalposts shifted mid-swing.
Judge the work
Critique projects, never people. First-timers with a real build beat veterans with slop, every time.
No refunds
You're buying a room, help and a real verdict. You get all three whichever colour your name ends up.