Custom domains

LitCurate does not hard-code a scientific domain. Start from the shipped template:

Starting point	When to use
configs/config.yaml + schemas/example/record.json + prompts/example/prompt.md	Complete template and dry-run starting point

The example record schema is a concrete list extraction (EOS V0/K0/Kp). Use it as a pattern for any domain that extracts repeated structured rows from papers.

To support a new topic:

1. Copy configs/config.yaml → e.g. configs/my_domain.yaml
2. Write the research user_goal and filter criteria
3. Copy schemas/example/record.json → schemas/my_domain/ and adapt fields
4. Copy prompts/example/prompt.md → prompts/my_domain/ and adapt instructions
5. Point extraction.schemas at those files (empty_list_field must match your list key)
6. Set LLM providers for search / filter / extract as needed
7. Run with export CONFIG=configs/my_domain.yaml

No Python changes are required for most new domains.

Downstream projects

LitCurate (this package)
    → runs/<id>/artifacts/output/database.json
        → Your app / DB / notebook

Keep experimental campaigns and large domain corpora outside the package repository when you publish LitCurate itself.

See also: Schemas & prompts, Export format.