nemotron-add-pattern

To add a reusable machine learning pattern, you create a pattern file with required frontmatter and body structure under the Nemotron step catalog, then regenerate the catalog for proper indexing. This keeps recurring training or evaluation decisions discoverable.

A reusable ML decision pattern encodes recurring training, evaluation, or optimization decisions into a structured format for future reference. You should create one when you notice the same decision, like a small-data LoRA choice, coming up repeatedly during planning.

You validate a new ML pattern by running the relevant tests to verify the pattern file, scope, and catalog output are correct. This validation workflow ensures the pattern metadata remains consistent with repository conventions and discoverable.

Yes, adding a pattern requires valid frontmatter to keep pattern metadata correct and discoverable. You must specify the pattern topic, triggers, affected steps, and confidence level so the catalog regeneration indexes the new guidance properly.

Yes, the pattern workflow applies to contributor workflows that add cross-cutting guidance for recurring training, evaluation, or optimization decisions. It maintains consistency with repository conventions while making the guidance easy to apply.

Encoding ML decisions as catalog patterns requires strict adherence to repository conventions, including valid frontmatter and test validation. Patterns are limited to cross-cutting guidance and require catalog regeneration to remain discoverable and correct.