kaxil opened a new pull request, #74707: URL: https://github.com/apache/airflow/pull/74707
`ClassifierRetryPolicy` is experimental because "what it does when the classifier is unsure or unavailable is changing". That change landed in #73501: the model call failing, or an answer under `min_confidence` or without a confidence, goes to `fallback_policy`, then `fallback_rules`, then the task's own retry settings. Since then the policy has only gained `include_traceback` (#74308), and nothing open changes that order. The policy works with any text model, so the decision-model install floors aren't a reason to keep it experimental either. This moves `ClassifierRetryPolicy` and `ErrorCategory` into the stable table. The promise: the model can only answer with a name from `categories`, and that name's `ErrorCategory` decides whether to retry and after what `delay`. `DEFAULT_CATEGORIES` keeps its seven names and their retry or fail split, since Dags already build on it with `**DEFAULT_CATEGORIES`. The fallback order above stays, and secrets are masked as for `LLMRetryPolicy`. Category descriptions and instructions are prompt text, which the page already leaves out of the promise. Only the `typesafe` extra stays experimental, because the `typesafe-sdk` it installs is pre-1.0. The note on the decision models page now says just that. The Why Common AI page (#74685) still called `ObjectStorageToolset`, per-tool approval and the batch operator experimental, though all three are in the stable table. This drops those labels too. #74694 rewrites the same experimental row for `ClassifierRetryPolicy` and the note on `decision_models.rst`; whichever lands second needs a small rebase. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
