sn-research-planning

To generate a deep research plan, provide your raw request as input. The system extracts goals and boundaries, then outputs a structured plan.json defining research dimensions, methods, and stop conditions without pre-writing conclusions.

Question decomposition in deep research planning breaks a broad query into independent, sub-report-able dimensions. Each dimension receives explicit dependencies, expected outputs, and stop criteria to guide downstream sub_reports/*.md creation.

Define report structure constraints by specifying the report shape within plan.json. You can optionally use report-format discovery to identify domain-specific conventions, compressing them into structural boundaries for the final report.md.

Yes, you can scope research boundaries and stop conditions before writing conclusions. The planning process extracts scope from request.md and defines sufficiency conditions for each dimension, ensuring the final report.md remains unbiased.

The best way to map search strategies is to generate an execution-ready plan.json. This plan assigns specific search strategies, methods, and execution orders to each independent research dimension for systematic sub-report generation.

Yes, you need a request.md file to start the research planning workflow. The system reads this input file to extract the research goal, audience use cases, and assumptions necessary to produce the actionable plan.json.