gh-xj/rd-deliberate

rd-deliberate: Decision Drafting

Evaluate alternatives and draft a D-NNN decision file based on assembled evidence.

When To Use

• Dispatched after rd-gather completes successfully
• Second worker in the research-decision chain: gather → deliberate → challenge → decide

Input

• Evidence assembly from rd-gather: synthesis/decisions/evidence/D-NNN-evidence.md
• Domain findings and analysis referenced in the evidence assembly

Output

• synthesis/decisions/drafts/D-NNN-draft.md

Responsibilities

1. Evaluate alternatives: Identify 2-3 viable approaches and build a trade-off matrix with decision-specific columns (not generic pros/cons)
2. Draft decision: Write a complete D-NNN decision file following the synthesis/decisions format
3. Use global D-NNN ID sequence: Check existing decisions across all domains and _cross-domain/decisions/ for the next available D-NNN ID
4. Set status to proposed: Decisions always start as proposed — human review required for accepted
5. Read existing decisions: Do NOT contradict finalized (accepted) decisions without explicit justification

Draft Decision Format

# Decision: D-NNN — [Title]
Date: YYYY-MM-DD
Status: proposed
Decision-makers: [to be confirmed by human review]

## Decision
[One paragraph: what we decided and why]

## Rationale
[Why this approach, citing specific findings by F-NNN ID]

## Alternatives Considered
| Alternative | Why rejected | Finding reference |
|-------------|-------------|-------------------|
| [Option A]  | [reason]    | F-NNN             |
| [Option B]  | [reason]    | F-NNN             |

## Approaches Evaluated
[Trade-off matrix with decision-specific columns — include when 2+ approaches were seriously evaluated]

## Evidence Basis
- Finding F-NNN: [summary of what it establishes]
- Finding F-NNN: [summary of what it establishes]

## Risks & Open Questions
[Known risks, uncertainties, conditions that could trigger revisiting]

## Implementation Notes
[Pointers to exp-job-hunter main repo or relevant domain directories, if applicable]

Trade-off Matrix Columns by Decision Type

Use decision-specific columns, not generic pros/cons:

• Platform scraping decisions: Columns: data richness, rate limit tolerance, ToS risk, maintenance burden, UK coverage
• Algorithm/tech decisions: Columns: accuracy, throughput, latency, infrastructure cost, implementation complexity
• Legal/compliance decisions: Columns: legal risk, ICO audit risk, user trust impact, implementation effort
• Infrastructure decisions: Columns: monthly cost at target scale, operational complexity, vendor lock-in, time to implement

Self-Check (Level 1 Self-Correction)

Before completing, verify:

• [ ] Every cited finding (F-NNN) exists and is referenced in the evidence assembly
• [ ] At least 2 alternatives are evaluated with specific rejection rationale
• [ ] No contradiction with existing accepted decisions (check synthesis/decisions/ and _cross-domain/decisions/)
• [ ] Decision paragraph is clear and actionable
• [ ] Trade-off matrix uses decision-specific columns (not generic)

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