We realized that article and item quality are special cases of a more general "content quality" concept.
We're not going to merge the existing models. Instead, were introducing a new "model family" property which can be used for autodiscovery of content quality models. Here's an outline of how clients should access the models:
- Most clients should will consume specific models, and these model names should be set in configuration. When making a request to the ORES API, always include the models parameter, e.g. models=editquality|articlequality.
- Some clients like an explorer UI may be able to take advantage of new models, without having the names hardcoded. These clients should make an initial request to list all models, e.g. https://ores.wikimedia.org/v3/scores/?model_info . A model_family field included in responses will help the client determine how each model can be used. Some experimental or deprecated models may be is_hidden.
- Other clients will request "all" models in order to pass the data through, e.g. to a generalized ORES cache. These should continue to do this without any changes. The server will hide the hidden models as needed. Clients shouldn't make any assumptions about what models are available in the "all" list.