Patterns

Patterns show how Dracon's primitives compose into real-world solutions. They assume you're comfortable with the basics from the tutorials.

Start Here

• Runtime Contracts - Keep runtime-only config declarative with !deferred, !require, and !assert.
• Constructor Slots - Let config choose types and builders with dynamic tags like !$(...).
• Layered Vocabularies - Build a reusable config language in layers with <<(<):.
• Hybrid Pipelines - Compose Python callables, YAML templates, and partials into one pipeline.
• Higher-Order Config - Use config to manufacture configured callables, not just data.

More Patterns

• Dynamic Skeleton - You have M datasets and N configs. Without composition, that's M*N files. With the skeleton pattern, it's M+N.
• Weighted Registries - Define the full list once, then make variants by turning a few items up, down, or off.
• Config Templates - Repeated config blocks that differ by a few parameters. Parameterize them once, stamp them out.
• Sweep Generation - Generate experiment grids with !each and expressions instead of writing them by hand.
• Composition Stack - Push, pop, and fork config layers at runtime for multi-phase pipelines.
• Anti-Patterns - Common mistakes and what to do instead.