How a Yard actually works.

Hackyard sets it by default. But the theme is also the thing a sponsor can set — a company or an indie backer can bring the constraint itself: a theme, a model, a tool, an API, or a challenge for the yard to build against.

That's deliberate. The constraint is the most valuable thing a sponsor can own, because it shapes all 50 builds instead of sitting in a banner nobody looks at. Sponsored Yards are also where prize pools come from — a backed Yard can put real money or resources behind the weekend on top of the standard badges and champion sponsorship.

Either way it reaches every builder at the same moment, and no sponsor gets a say in who wins — the vote is the vote. Sponsor a Yard →

• Repo URL — required
• Writeup — required, 500 characters max
• Model declaration — required, and published on your submission page
• Demo video — optional but strongly recommended (YouTube, Vimeo, or Loom play inline on your submission page)
• Screenshot — optional, and it's what people see first on your card

Yes. Dependencies, frameworks, and CLI scaffolds are fine. Your own pre-written project code is not. Running create-next-app, installing packages, or starting from a public boilerplate is normal work, not a head start.

The real question is: did you start from something you built earlier? That's the only version of this that gives anyone an unfair edge, and it's exactly what the repo timing panel measures.

So don't burn two of your 48 hours hand-rolling something you could install. “All code written during the 48 hours” means the code that makes your project yours.

Every submission page shows its repo's timing panel, pulled live from GitHub: when the repo was created, how many commits it has, and the span from first to last commit — next to the Yard's build window.

If a repo was created before kickoff, the panel says so, in public. We don't remove anyone for it; we show the dates and let the yard weigh them. Repo creation dates come from GitHub and cannot be edited by the author. Commit dates can be rewritten, so the panel labels them as context, not proof.

Check the Yard — the rule is set per Yard, not once for all of Hackyard.

Yard #1 names one: Grok 4.5, via Cursor, the xAI API, or the grok.com app. For that weekend everyone builds on the same model, so nobody wins by out-spending anyone.

If a Yard doesn't name one, it's open-model — bring Claude, GPT, Grok, Gemini, something local, whatever you've already got — and the theme is the only constraint that matters. Expect that to be the common case going forward.

No — and here's the honest reason why. Naming a model doesn't just set a constraint, it sets a bill. If you already pay for a $20 Claude plan, a Grok-only weekend asks you to buy a second subscription to join. A lot of good builders will quietly pass on that, and we'd never hear from them.

That's a bad trade for a community this small. So a named model is the exception — usually because a sponsor brought that model as the Yard's constraint, or because the point of that particular weekend is seeing one model in 50 different hands.

Expect most Yards to be open-model: bring Claude, GPT, Grok, Gemini, a local Llama, or no AI at all, and let the theme do the constraining. Everyone shows up with what they already have.

There's no free tier that reaches it — the free grok.com and X tiers don't include Grok 4-class models. But you do not need the $300/mo plan, which is the assumption that scares people off. Two cheap routes:

• Cursor Pro — $20/mo. Grok 4.5 is in Cursor's included model pool on every paid plan, and it was tuned with Cursor for long-running coding work. If you already pay for Cursor, you're done — nothing to buy.
• The xAI API — pay as you go. About $2 per million input tokens and $6 per million output, no subscription and no minimum. xAI has run signup credits for new accounts; check their current terms, because a 48-hour build can land inside them and cost you nothing.

So the real range for entering Yard #1 is roughly $0 to $20, not a $300 subscription. Prices change and we don't control them — check xAI and Cursor before you buy anything.

Yes. The named model applies to code generation. Everything else is fair game. Copilot-style autocomplete in your editor, pasting a stack trace into whatever chatbot you like, generating a logo or sound effect, transcribing your demo voiceover — none of that counts against the model rule.

The line is the one in the checkbox you tick when you submit: the named model should be your primary code-generating model. Nobody is expected to uninstall their tooling for a weekend, and a rule that strict would only punish whoever read it most honestly.

First, check what it actually costs — people assume a named model means an expensive subscription, and for Yard #1 that's wrong. Grok 4.5 comes with Cursor Pro at $20/mo, or through the xAI API pay-as-you-go with signup credits that can cover a whole weekend. See the cost breakdown above before you count yourself out.

If it still doesn't work for you, that's a real cost we're not going to talk around, and sitting one out is a completely reasonable call. You can still vote in it, still claim a spot in future Yards, and still get the announcement emails — and the next Yard is far more likely to be open-model precisely so this stops being a question. If you're holding a spot you can't use, releasing it is the most useful thing you can do; it goes straight to the next person on the waitlist.

Straight answer: we can't detect it, and we won't pretend to. Nothing in a repo reveals which model wrote the code. There's no watermark, and “AI detection” tools can't reliably tell human from AI, let alone one model from another. Anyone claiming to verify this is guessing.

What we do instead is make the claim public. Every submission carries a model declaration — which model, which surface, under your handle — and the database will not accept a submission without one. We enforce that a declaration exists. Whether it's true is on you.

A receipt is an optional link showing your model session: a shared Grok conversation, your xAI console usage page, or a session log committed in your repo. Attach one and your card shows a Receipts marker during voting.

You do not need one, and skipping it is not held against you. Most builders won't bother, and that's fine. A receipt shows a session existed — it does not prove that model wrote what shipped, and we never label it as verification. The strongest thing it says is “this builder showed their work.”

Honestly? Mostly the fact that it wouldn't buy them much. You don't win a Yard on code quality — you win it on a peer vote for what you shipped and how you demoed it. A marginally stronger model moves that very little. A two-week head start moves it enormously, which is exactly the one we publish on every submission.

And the shared model is the whole reason a Yard means anything. Swap it and you didn't beat 49 people under the same constraint — you did something else, by yourself, and called it a win. We enforce what's checkable and stay honest about the rest.

Your call — it's your one vote. But the things worth weighing:

• Does it fit the theme? Everyone got the same one.
• Does it actually work? A demo video beats a description.
• Did they ship something real in 48 hours, under the same constraints you had?
• Context, not verdicts: the repo timing and model panels are there to inform your read, not to hand you a ruling.

This one matters, so here's the full answer.

Yes, position is an advantage — the first card in a grid always gets more eyes than the last. If we listed submissions in the order they arrived, whoever submitted earliest would get a permanent, unearned edge, and “every project gets seen” would be a slogan rather than a fact.

So during voting, the order is shuffled — and reshuffled continuously. Every time the page is loaded fresh, submissions are randomly reordered. Nobody holds the top-left slot; over a voting window, every build rotates through the good positions and the bad ones roughly equally.

Once voting closes, the shuffle stops and submissions are ranked by vote count, because at that point the page is showing results rather than asking for votes.

• Champion badge — permanent, on your profile
• Featured on the home page — your build, pinned as Reigning Champion
• A published writeup on what you made and how
• The big one: a 1-year sponsorship from Hackyard — 365 days pinned to the home page, boosted on our X, first dibs on every future Yard, and a permanent Champion badge

• Champion — 1st place
• Top 3 — 2nd and 3rd
• Shipped — everyone who submitted