jicaiinc/golemancy/research-deep

Deep Research Execution

Execute deep research on individual items from a research outline, producing structured findings for each item.

Workflow

Step 1: Load Research Outline

Read the research outline produced in the first phase:

• Items list with descriptions
• Field definitions with detail levels
• Execution configuration

Step 2: Resume Check

Check for previously completed research items. Skip already-completed items to support incremental research.

Step 3: Batch Execution

For each research item:

1. Search the web for comprehensive information
2. Fill in all defined fields from the outline
3. Mark uncertain values with [uncertain] tag
4. List all uncertain fields at the end
5. Validate completeness against field definitions

Step 4: Monitor Progress

After each batch:

• Report completion status
• Surface any items requiring manual verification
• Proceed to next batch with user approval

Step 5: Summary Report

After all items are researched:

• Total completion count
• Items with uncertain fields flagged
• Key findings and patterns observed
• Recommendations for follow-up

Research Quality Standards

• Every factual claim must have a source URL
• Distinguish between verified facts and inferences
• Use multiple independent sources for critical data points
• Record the date of information retrieval
• Flag contradictory information from different sources

Output Format

For each research item, produce structured data containing:

• All fields defined in the research outline
• Source URLs for each data point
• Confidence level (verified / likely / uncertain)
• Date of last verification
• Notes on conflicting information

Best Practices

• Prioritize primary sources over aggregators
• Cross-reference data across at least 2 sources
• Note when information may be outdated
• Include both quantitative and qualitative findings
• Maintain consistent formatting across all items