Contribute to the library.

Registering puts your toolkit in the library's directory, with a landing page of its own. It is an agreement in principle: nothing in it is binding, and you don't need it to submit a solution.

1. 1

   Open the form

   Fill in Register a toolkit in the working group's tracker, one issue per toolkit.

2. 2

   Add your links

   Documentation, installation, tutorials and where users ask for help. Your landing page is built from them.

3. 3

   Keep it current

   Edit the issue whenever the details change.

A solution is a runnable implementation of one of the library's models. Keep the model and its calibration fixed; the method, the prose and the format are yours. Submit on behalf of a toolkit's project or as an individual.

1. 1

   Read the model

   Start with its description and submission guide, such as buffer stock's.

2. 2

   Solve it

   A notebook, a MyST document or another runnable format, reporting the results the description lists.

3. 3

   Add metadata.yml

   A short file: the authors and maintainers, the description version you follow, and your AI disclosure.

4. 4

   Open a pull request

   Add your directory under solutions/<model>/. A check names any missing field.

Your solution stays yours. You author and maintain it, and changes to it, by anyone, need a maintainer's approval. How submissions are reviewed is being settled in #6.

A model description is the template every solution of a model follows: the model, its calibration and the results each solution reports. The baseline set is a starting proposal, not a ceiling. Proposing a model is for domain experts: people who know the model and its literature well enough to state it precisely.

The specification is the contribution.

Solutions from different toolkits can be compared only when they solve the same problem. A gap in the description lets correct solutions disagree for reasons no one can trace, and every solution inherits it. So the library asks more of a description than of any one solution.

A good description is

Complete

Someone who hasn't read the source paper can implement the model from the description alone.

Unambiguous

Two people implementing it independently solve the same problem, with any free choice named.

Correct

Checked against its sources, with any departure from the paper stated and explained.

Comparable

Results are evaluated at fixed points, so correct solutions by any method agree.

Solvable by more than one toolkit

Nothing in it relies on a feature that only one toolkit has.

A baseline

A standard version of a model the field knows, kept as small as its purpose allows.

What it pins down

• Sources
• Model
• Calibration
• Conditions
• Numerical solution
• Results
• Checks

The buffer stock description states each of these and is the example to follow.

Each model description has its own version number, so every solution can say exactly which one it follows.

Each version has an entry in the model's changelog, and is tagged and released with its PDF.

Every description and solution says how much AI helped produce it and whether a person has verified it. Both describe the process: neither claims that a document is free of errors, and neither ranks documents, toolkits or submitters.

• Human AINo AINo AI was used.
• Human AIAI-editedPeople wrote it; AI suggested edits.
• Human AIAI-assistedAI drafted some or most of it; people edited it throughout.
• Human AIAI-generatedAI drafted it; people edited it lightly.

Report the highest category that applies to any substantial part, such as one section or the code. Choices about the model, the calibration and the results always belong to the authors.