Contribute to T2C-Registry

Contributions are welcome through GitHub issues:

Propose a resource

Requirements

• benchmark has to be hosted in a public, version-controlled repository;
• include, where possible, the code and configuration files required to reproduce the reported results;
• identify the dataset, data split, graph snapshot, and database version used for evaluation;
• clearly distinguish between LLM-assisted and manually curated data;
• remain open to corrections, extensions, and reuse by the community.

Fastest way to get included

The quickest path to onboarding is to send a single JSON file: a flat list of entries, each with a question, a cypher query, and optional notes (e.g. the query's complexity level).

[
  {
    "question": "Which genes are associated with breast cancer?",
    "cypher": "MATCH (g:Gene)-[:ASSOCIATED_WITH]->(d:Disease {name: 'breast cancer'}) RETURN g.name",
    "notes": "complexity: simple, single hop"
  },
  {
    "question": "How many pathways involve genes on chromosome 17?",
    "cypher": "MATCH (g:Gene {chromosome: '17'})-[:PART_OF]->(p:Pathway) RETURN count(DISTINCT p)",
    "notes": "complexity: medium, aggregation"
  }
]

Together with the JSON file, share the Neo4j connection details for the graph the queries run against:

• the bolt endpoint (e.g. neo4j+s://host:7687);
• the database name;
• username and password (a read-only account is preferred).

These map directly onto the endpoint block of a databases entry in docs/registry.yaml, so once received, adding the resource is a matter of registering the connection and dropping in the JSON file.