Schema validation now works across Python and TypeScript

The tracer bullet hit its first real milestone this week: a GeoJSON feature validated in Python, serialised to JSON, deserialised in TypeScript, and validated again — with both ends agreeing on the schema.

This sounds mundane, but it’s the foundation everything else builds on. When debrief-io transforms a REP file into GeoJSON, and debrief-stac stores it, and a VS Code extension displays it — they all need to agree on what a “track” looks like.

The approach

LinkML gives us a single source of truth. The schema is defined once:

classes:
  Track:
    description: A vessel track with positions over time
    attributes:
      id:
        range: string
        required: true
      name:
        range: string
      positions:
        range: Position
        multivalued: true

From this, we generate:

• Pydantic models for Python
• JSON Schema for validation
• TypeScript interfaces for the frontend

What’s next

The schema is minimal right now — just enough to prove the pipeline works. Next steps:

• Add sensor data types
• Define the STAC Item structure for plots
• Get the REP file parser producing valid output

→ See the code