Making open-source improvement cumulative
Open-source AI already produces many models, adapters, prompts, datasets, and tools. Model breeding can make that activity more cumulative by standardizing how packages declare compatibility, evaluation evidence, lineage, and operating costs.
The benefit is a healthier commons. Contributors can improve a skill package without needing to control the whole stack. Maintainers can accept or reject descendants based on published tests instead of reputation alone.
Commons primitives
| Primitive | Purpose |
|---|---|
| Contract schema | Makes packages interchangeable. |
| Evaluation suite | Makes improvements comparable. |
| Lineage record | Makes inheritance visible. |
| Merge recipe | Makes recombination reproducible. |
| License metadata | Makes reuse decisions explicit. |
FUNCTION accept_open_source_descendant(package, project)
REQUIRE package.contract.version IN project.supported_contracts
REQUIRE package.lineage.parent_hashes_present
REQUIRE package.evaluation.reproducible
REQUIRE package.license.compatible_with(project.license_policy)
RETURN add_to_candidate_registry(package)
END FUNCTIONPositive norm
Publish the tests, not only the weights. A model ecology grows faster when every useful artifact arrives with evidence that others can repeat.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.