Marketplace

The marketplace at /marketplace is the discovery layer. Anything published — yours or someone else’s — shows up as a tile users can install or fork into their own projects.

Browsing

The filter row at the top lets you slice the listings by category:

• All
• Regression
• Classification
• Time series
• Probabilistic
• Causal / discovery
• Optimization

Each tile shows:

• Icon + title
• Category chip
• One-line description
• Paper citation if applicable
• Author / org
• Listing type (Prior, Model, or Brain)

Click a tile to expand a card with the full description, code/architecture preview, and the primary action button (Install or Fork).

Install vs Fork

	Install	Fork
What you get	A copy of the listing inside an existing project	A brand-new project seeded from the listing
When to use	You already have a project and want to add this prior/model to it	You’re starting fresh — typically for paper reproductions
Edit afterwards	Yes	Yes — it’s your project
Counts toward the original’s fork count	No	Yes

Install flow (prior or model)

1. Click Install on a tile
2. Modal opens with:
   • The listing summary
   • A target project dropdown (your own projects, fetched via /projects)
3. (Model installs only) A compat-chip appears below the project picker:
   • 🟢 Green — “Compatible with N matching <category> prior(s) in this project: <names>”
   • 🟡 Amber, no priors — “This project has no priors yet. You’ll need a <category> prior installed before this model can train.”
   • 🟡 Amber, no match — “No <category> priors in this project (N other priors present). The model will install but won’t have a matching prior to train on.”
4. Click Install prior → / Install model → to commit

The installed listing appears in the project’s Priors / Models tab. From there you can edit it, attach it to a new Run, or fork it further.

Fork flow

1. Click Fork on a tile
2. Modal opens with:
   • The listing summary
   • A project name input (pre-filled with the listing’s title)
   • A description textarea
3. Click Fork prior → / Fork brain →

A new project is created with the listing copied in as a seed prior/model/eval/run. The original’s forkCount increments. You land on the new project’s overview page.

Forking a paper-backed brain

If you fork a brain whose prior is paper-backed, the project lands with:

• The paper’s prior code verbatim
• The paper’s model architecture
• The paper’s eval suite
• The paper’s hyperparameters (learning rate, batch size, step count, seed)
• The paper’s preamble as the project readme

You can train it as-is (reproducing the paper) or edit any artifact to diverge. See Paper reproductions for the full reproduction-vs-diverge picture.

Compat-chip details (model installs)

The compat-chip is a guardrail. Models in the marketplace declare a category (Regression / Classification / Time series / etc.). When you install one into a project, the chip fetches the project’s existing priors and checks whether any share that category.

The chip is informational — install proceeds either way. But it surfaces the next step the user needs to take:

• Match found → ready to train
• No match → install a prior of the right category before the model is usable
• No priors at all → fork or install a prior first

Publishing your own

Not yet exposed in basic mode. From developer mode:

1. Open the project you want to publish from
2. Go to Priors (or Models, or the project root for a brain)
3. Click Publish to marketplace on the artifact
4. Fill in: title, description, category, paper citation (optional), license

Your listing appears in the marketplace immediately for everyone else to discover. Publishing is org-scoped — you control which org’s listings go public.

Curated paper reproductions

Some listings are curated paper reproductions — the marketplace adds a ● Reproducible study badge linking back to the original repo (e.g. ifBO’s GitHub). Forking these creates a project pre-loaded with the paper’s full bundle; the project shell’s left sidebar shows the same badge once forked.