Persistent instructions, or a portable falsifiable model

A Claude Project holds instructions and files that persist across chats. A model does something narrower and stranger: it states a mechanism, what would refute it, and when its author gives up on it.

The short answer

Project instructions shape how an assistant behaves. A model states a claim about the world that can be graded — and moves to any assistant, unchanged.

Overlap is real: both put standing context in front of a model. The difference is what that context is for.

Project instructions are configuration — tone, format, what to remember, which files to consult. They are not claims about anything, so there is nothing to be right or wrong about.

A garden model is a claim. It says what drives outcomes in some domain, what observation would falsify that, and the conditions under which the author will retire it. That makes it gradeable, and it makes an author accountable in a way a system prompt never is.

The second difference is portability. A Project lives in one product. A model is a git repo — paste it into Claude, ChatGPT, or Gemini, or clone it and run it against a local model on your own machine. Same document, no lock-in.

Side by side

Claude Projects	Model Meets Reality
Configures how the assistant behaves	States what is true and what would refute it
Lives inside one product	A git repo; runs in any assistant or offline
Nothing to grade	Claims registered with criteria frozen in advance
Private to your account	Public by default, so others can disagree with it
Files persist across chats	Self-contained — needs no memory or prior chat

What Claude Projects does better

For actually getting work done, Projects are more useful more often: persistent files, real memory, and no need to paste anything. If you want an assistant configured your way, use a Project. Use a model when you want the reasoning to be a claim someone can check, including you, later.

Where this stands today

Said plainly, because the whole point of this site is not overclaiming. The Model Garden is opening small — no models listed as of September 2026. Every record shown is self-graded: the author's own count of claims made and resolved, labelled as such on each card. Nothing here is ranked, and no independent resolution layer exists yet. Claude Projects has things this does not, named above rather than omitted. Compare the designs, not the scoreboards — there is no scoreboard.

https://github.com/someone/their-model Help me use this

Paste that into any assistant. It reads the model — premises, falsifiers, deletion clause — and reasons through it.

The difference that is not on the table

Claude Projects answers your question. It does not keep a record of whether the answer held. That is not a flaw — it is not what an assistant is for. But if you have watched one thing closely for years, that knowledge is already a model: premises, predictions, blind spots. Writing it down turns it into something an assistant can reason through, and something a date can settle.

Share your understanding →

Browse the Model Garden Publish your own