Thinking about recipe formats more than anyone should

Original Article ↗ Hacker News Discussion ↗

Data formats: JSON, XML, plain text, DSLs

  • Strong disagreement over using JSON for rich text recipes; some prefer XML-like markup, others find complex XML harder to parse than well-constrained JSON.
  • Plain text (or markdown/AsciiDoc) is praised as easiest to write and read, but criticized as poor for reliable machine parsing at scale.
  • Several custom formats are mentioned (Cooklang, recipe-lang, custom YAML/TOML, BatchML, Google’s recipe schema). Some find them elegant; others think average users won’t learn a DSL and just want a text box.

What a recipe “really is” (lists, trees, DAGs, workflows)

  • Multiple models proposed: linear list, upside-down tree, DAG, project plan, generalized checklist.
  • Tree/DAG views help reason about ingredient reuse, intermediate products, parallel steps, and multi-component dishes.
  • Skeptics argue real cooking knowledge (timing, conditions, adjustments) is richer than any neat graph representation.

Presentation formats & UX

  • High value placed on presentation: clear flow, prep vs cooking, reuse of ingredients, and avoiding being “meanwhiled” by hidden parallel steps.
  • Popular visual ideas:
    • Multi-column tables (ingredient/quantity/instruction) and “Cooking for Engineers”-style grids.
    • Gantt charts and time–resource diagrams, especially for coordinating multiple dishes or holidays.
    • Tree/graph visualizations that show how subcomponents combine.
  • Phone UX issues: hard to see ingredients and steps simultaneously; some tools inline or reveal quantities within steps.

Role of LLMs

  • Enthusiasts: with modern LLMs, authoring can stay free-form; machines can generate whatever structured JSON/XML/graph is needed.
  • Skeptics: current LLMs are not yet trustworthy for high-stakes aggregation (e.g., bulk ingredient accounting for catering) or safety-critical details.

Over-engineering vs “just cook”

  • Some see heavy modeling as pseudo-engineering that ignores how professionals actually cook (simple ratios, minimal instructions, lots of tacit skill).
  • Others enjoy the modeling for planning, shopping lists, scaling, and exploring variation space (e.g., curry combinatorics).
  • There’s recurring tension between human-friendly, informal recipes and machine-friendly, highly structured systems.