gather

To collect research findings with sources and confidence levels, use a structured investigation workflow that searches code first, then the web, returning numbered findings with evidence summaries and confidence ratings.

The best way to investigate an unfamiliar codebase feature is to run a code-first discovery workflow that maps the implementation, follows up with web context, and outputs cited findings with assigned confidence levels.

Structured research with confidence ratings works by searching local code and web sources, summarizing the evidence into concise numbered findings, and labeling each finding with a confidence score to show result reliability.

Yes, you can expand prior research findings when investigating a bug by applying a discovery workflow that builds on existing notes, searches for new code and web evidence, and refines the output with updated sources and confidence labels.

Codebase investigation requires web search for broader context only after an initial code-first search fails to provide sufficient evidence, ensuring the final structured findings include comprehensive cited sources and confidence ratings.

You should not use an open-ended research workflow when you need a single definitive answer rather than assembled evidence, because this process produces structured findings with confidence levels meant for distillation rather than immediate final conclusions.