plinde Socratic Debate Framework
A structured approach to deliberation using multiple AI perspectives to stress-test ideas and reach well-reasoned conclusions.
When to Use
- Evaluating whether to accept or reject a proposal (PR feedback, RFC, design decision)
- Making architectural or technology choices with significant tradeoffs
- Deciding whether something is worth the effort/complexity
- Any situation where "it depends" is the initial answer
The Four Perspectives
1. Advocate FOR (Proponent)
Role: Make the strongest possible case in favor of the position.
Mindset:
- Assume the proposal has merit and find the best reasons why
- Consider benefits that may not be immediately obvious
- Think about precedent, standards, and long-term implications
- Acknowledge weaknesses only if doing so strengthens credibility
Output format: 250-400 word persuasive argument with a memorable closing line.
2. Advocate AGAINST (Devil's Advocate)
Role: Stress-test the idea by making the strongest counterargument.
Mindset:
- Look for hidden costs, complexity, or unintended consequences
- Question whether the problem being solved is overstated
- Identify alternative approaches that might be simpler
- Consider opportunity cost and what else could be done instead
Output format: 250-400 word counterargument with a memorable closing line.
Important: The goal is constructive challenge, not dismissal. A good devil's advocate helps strengthen ideas.
3. Neutral Analyst
Role: Objectively weigh both sides and identify the key tradeoffs.
Mindset:
- Remain impartial while still being willing to draw conclusions
- Identify where the debaters agree (often more than expected)
- Surface context or constraints that affect the decision
- Consider hybrid approaches or middle grounds
Output format:
- Balanced analysis (200 words)
- Tradeoffs table (if applicable)
- Preliminary verdict with confidence level (low/medium/high)
4. Scribe/Moderator
Role: Synthesize all perspectives into a coherent summary and final verdict.
Responsibilities:
- Extract the most compelling points from each perspective
- Identify areas of agreement across all three debaters
- Resolve conflicts by weighing evidence quality
- Deliver a clear, actionable recommendation
Output format:
## Socratic Debate Summary
### Topic
[Original question/topic]
### FOR (Proponent)
[Key points summarized]
### AGAINST (Devil's Advocate)
[Key points summarized]
### NEUTRAL (Analyst)
[Key tradeoffs identified]
### Points of Agreement
[What all perspectives agreed on]
### Moderator's Verdict
**Recommendation:** [accept/reject/modify/defer]
**Confidence:** [low/medium/high]
**Key Factor:** [The decisive consideration]
### Suggested Next Steps
[If applicable]
Debate Principles
Intellectual Honesty
- Debaters should make their best arguments, not strawmen
- Acknowledge strong points from other perspectives
- Distinguish between facts, interpretations, and opinions
Steelmanning
- Each perspective should address the strongest version of opposing arguments
- Avoid attacking weak or irrelevant points
- Give credit where due
Actionable Outcomes
- The goal is a decision, not endless deliberation
- Every debate should end with a clear recommendation
- Include confidence levels to indicate certainty
Appropriate Scope
- Focus on the specific question at hand
- Avoid scope creep into tangential issues
- Time-box arguments to maintain focus
Example Debate Topics
| Category | Example Topic |
|---|---|
| Code Review | "Should we require this change before merging?" |
| Architecture | "Should we adopt microservices for this system?" |
| Process | "Is the added ceremony of RFCs worth it for this team?" |
| Technology | "Should we migrate from X to Y?" |
| Prioritization | "Is this bug fix more urgent than feature work?" |
Anti-Patterns to Avoid
- False Balance - Not all positions deserve equal weight; evidence matters
- Analysis Paralysis - The goal is a decision, not perpetual debate
- Bikeshedding - Don't let low-stakes issues consume debate resources
- Motivated Reasoning - Debaters should follow evidence, not justify predetermined conclusions
- Ignoring Context - Recommendations should account for real-world constraints
Output Behavior
Default: Local reasoning, silent action
By default, the debate reasoning is shown locally to the user (full FOR/AGAINST/ANALYST/MODERATOR output in the conversation), but consensus fixes are applied silently — no debate summary is posted to PRs, issues, or external systems. The agent goes straight to implementing the consensus recommendation.
This means:
- The user sees the full deliberation process in their terminal
- External artifacts (PRs, commits, comments) reflect only the outcome, not the process
- Commit messages and PR descriptions should describe what changed, not why the debate concluded X
Override: Post reasoning externally
If the user explicitly asks to share the debate reasoning (e.g., "post the debate to the PR", "include the reasoning in the ADR"), then include the Moderator's Summary in the external artifact.
Integration with Workflows
The Socratic debate format works well for:
- Code/skill review - Debate quality, then apply consensus fixes directly
- PR Reviews - Debate review feedback, then post only actionable comments (not the debate itself)
- ADRs - Include debate summary in "Considered Alternatives" only if explicitly requested
- RFCs - Use as structured feedback before approval
- Retrospectives - Debate proposed process changes
- Incident Reviews - Evaluate proposed preventive measures