gh-xj/rd-gather

rd-gather: Evidence Assembly for Decisions

Assemble and assess evidence from domain findings to prepare for decision drafting.

When To Use

• Dispatched by case-dispatcher when a research-decision case (D-NNN) is opened
• First worker in the research-decision chain: gather → deliberate → challenge → decide

Input

• D-NNN case record (topic, scope, related questions)
• Domain findings (F-NNN files) from multiple answered questions
• Cross-domain findings when decision spans multiple domains

Output

• synthesis/decisions/evidence/D-NNN-evidence.md

Responsibilities

1. Identify relevant findings: Read findings from questions related to the decision topic (same domain or cross-domain via _cross-domain/findings/)
2. Assess confidence sufficiency: For each finding, check confidence level (high/medium/low) and whether it is adequate for decision-making
3. Flag evidence gaps: Identify where more evidence is needed before a sound decision can be made — specific sub-questions unanswered, findings at low confidence, missing comparisons
4. Assemble evidence document: Write a structured evidence assembly to synthesis/decisions/evidence/D-NNN-evidence.md

Evidence Assembly Format

# Evidence Assembly: D-NNN — [Decision Topic]
Date: YYYY-MM-DD
Status: complete | gaps-identified
Related questions: Q-NNN, Q-NNN, ...

## Relevant Findings

| Finding | Question | Confidence | Key Claim | Sufficient? |
|---------|----------|------------|-----------|-------------|
| F-001   | Q-003    | high       | ...       | yes         |
| F-004   | Q-007    | low        | ...       | no — needs corroboration |

## Evidence Gaps

- [Gap description — what is missing and why it matters for this decision]

## Cross-Domain Evidence

- [References to findings from other domains, if applicable]

## Sufficiency Assessment

[Overall assessment: is the evidence base adequate for decision drafting, or should specific questions be investigated further first?]

Cross-Domain Considerations

Many decisions in the job-hunter SaaS span multiple domains. Common cross-domain evidence needs:

• Scraping + Legal: Evidence from both 01-scraping-acquisition and 05-gdpr-legal required for decisions about data collection methods
• Visa + Market: Evidence from both 03-visa-sponsorship and 04-uk-job-market for decisions about feature prioritisation
• AI + Infrastructure: Evidence from both 07-ai-integration and 08-infrastructure for decisions about ML pipeline architecture
• Dedup + Search: Evidence from both 02-dedup-data-quality and 06-search-ranking for decisions about index construction

Self-Check (Level 1 Self-Correction)

Before completing, verify:

• [ ] Every finding cited exists in the domain's synthesis/findings/ or _cross-domain/findings/
• [ ] Every finding references valid F-NNN IDs
• [ ] Gaps are specific and actionable (not vague)
• [ ] Cross-domain evidence has been checked where the decision spans domains

Max 2 self-correction iterations. If self-check still fails after 2 retries, emit status-recommendation: blocked with description.

Failure Path

If evidence is fundamentally insufficient (no high-confidence findings, critical gaps):

• Emit status-recommendation: blocked with description of what evidence is missing
• Dispatcher keeps D-NNN status unchanged and logs blockage
• Case remains open for human intervention or additional research questions