Skip to main content
Use the data-factory flow when the run should create or refine datasets, examples, labels, or research inputs before downstream work.

​
Goal

Start with clear intake criteria, generation constraints, review policy, and publish expectations.

​
MCP path

Ask your MCP client: 
Start a Managed Research project run for a data-factory workflow. Use directed effort, provider openrouter, and require an artifact manifest plus final report before publish.
The primary MCP path is the normal project-run launch path: create or select a project, attach the relevant repo or files, preflight, then call smr_trigger_run. For isolated work, use smr_start_one_off_run.

​
Good instructions

  • describe the target dataset or artifact
  • specify accepted source material
  • specify validation or review criteria
  • require rejected-example notes
  • require artifact manifest and publish summary

​
Expected evidence

  • generated or revised data files
  • review notes
  • rejected or deferred examples
  • final report
  • usage summary

​
Failure notes

Use project-scoped context for data-factory work. One-off runs are better for isolated repo tasks.