wyattowalsh/panel-debate-skill
panel-debate
VerifiedHosts interactive expert panel discussions on any topic. Dynamically generates diverse master-level expert personas, facilitates structured multi-round debate using Hegelian dialectic patterns, and synthesizes actionable consensus. Useful for exploring complex topics, making architectural decisions, evaluating trade-offs, or seeking diverse expert perspectives on strategic questions.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.3 KB |
| assets/.gitkeep | other | 117 B |
| examples/README.md | markdown | 1.1 KB |
| examples/architecture-decision.md | markdown | 7.3 KB |
| examples/business-strategy.md | markdown | 4.5 KB |
| examples/security-implementation.md | markdown | 2.8 KB |
| references/expert-generation.md | markdown | 28.8 KB |
| references/output-formats.md | markdown | 35.3 KB |
| references/research-foundations.md | markdown | 11.3 KB |
| references/synthesis-patterns.md | markdown | 32.1 KB |
| references/turn-taking.md | markdown | 27.1 KB |
references/synthesis-patterns.md
View on GitHub# Synthesis Patterns Reference
## Contents
- [Research Foundations](#research-foundations)
- [1. Hegelian Dialectic Fundamentals](#1-hegelian-dialectic-fundamentals)
- [2. Per-Round Synthesis Format](#2-per-round-synthesis-format)
- [3. Synthesis Generation Algorithm](#3-synthesis-generation-algorithm)
- [4. Final Consensus Generation](#4-final-consensus-generation)
- [5. Handling Unresolved Tensions](#5-handling-unresolved-tensions)
- [6. Templates for Synthesis Statements](#6-templates-for-synthesis-statements)
- [7. Examples: Raw Discussion to Synthesis](#7-examples-raw-discussion--synthesis)
- [8. Quality Checklist for Syntheses](#8-quality-checklist-for-syntheses)
- [9. Anti-Patterns and Failure Modes](#9-anti-patterns-and-failure-modes)
- [10. Meta-Synthesis: When the Process Itself Provides Insight](#10-meta-synthesis-when-the-process-itself-provides-insight)
- [Conclusion](#conclusion)
This document provides patterns, algorithms, and templates for generating meaningful syntheses from expert panel discussions using Hegelian dialectic principles.
---
## Research Foundations
Synthesis mechanisms are critical to multi-agent discussion effectiveness:
- **Self-Consistency via Discussion (PanelGPT)** `[Research-Backed]`: Framing consensus as discussion rather than voting improves outcomes—the discussion metaphor encourages reasoning, not just aggregation
- **Argument-Quality Weighting** `[Research-Backed: CISC (ACL 2025) - confidence weighting]`: Research shows majority voting can entrench initial errors due to LLM conformity; weighting by reasoning quality counteracts this
- **Explicit Deliberation** `[Research-Backed]`: Requiring agents to explicitly agree/disagree and justify produces better syntheses than implicit aggregation
- **Context-Dependent Conclusions** `[Research-Backed]`: Research indicates forcing binary consensus reduces accuracy; acknowledging "it depends" with specific conditions is more accurate
---
## 1. Hegelian Dialectic Fundamentals
### Core Structure
**Thesis (Initial Position)**
- A claim, position, or proposed solution
- Grounded in specific reasoning and evidence
- Represents one valid perspective on the issue
**Antithesis (Opposition/Challenge)**
- A competing claim that challenges the thesis
- Not mere negation, but an alternative perspective
- Reveals limitations or blind spots in the thesis
- Also grounded in evidence and reasoning
**Synthesis (Higher-Order Integration)**
- Preserves valid elements from both thesis and antithesis
- Transcends simple compromise or averaging
- Generates NEW understanding not present in either original position
- Resolves contradictions at a higher level of abstraction
- May reveal that both positions are context-dependent
### Key Principles
1. **Aufhebung (Sublation)**: Each synthesis both cancels AND preserves elements of thesis/antithesis
2. **Emergence**: Synthesis contains insights that neither position held individually
3. **Context Sensitivity**: Recognition that different contexts may favor different positions
4. **Progressive Refinement**: Each round's synthesis becomes next round's thesis
### Anti-Patterns to Avoid
- **False Balance**: "Both sides have good points" without integration
- **Premature Consensus**: Glossing over real disagreements
- **Abstract Handwaving**: Vague synthesis that lacks actionable content
- **False Dichotomy**: Presenting issues as binary when they're multidimensional
- **Wishy-Washy Hedging**: "It depends" without specifying what it depends on
---
## 2. Per-Round Synthesis Format
> [Implementation Choice] The four-category structure below (Agreements, Productive Tensions, Open Questions, Emerging Insights) is a design pattern, not research-prescribed. The categories operationalize research findings about explicit deliberation and context-dependent conclusions.
### Standard Template
```markdown
📋 Round N Synthesis:
**Agreements:**
• [Explicit agreement with same reasoning]
• [Implicit agreement with different paths to same conclusion]
**Productive Tensions:**
• [Key disagreement #1 with specific nature of divergence]
• [Key disagreement #2 with underlying assumption causing split]
**Open Questions:**
• [Unresolved empirical question requiring evidence]
• [Definitional ambiguity requiring clarification]
• [Value-based tension requiring stakeholder input]
**Emerging Insights:**
• [Novel understanding that emerged from the exchange]
• [Reframing that clarifies the true nature of the problem]
• [Hidden assumption now made explicit]
```
### Rich Template (For Complex Discussions)
```markdown
📋 Round N Synthesis:
**Strong Consensus:**
• [Point of unanimous agreement]
- Expert A's reasoning: [brief]
- Expert B's reasoning: [brief]
- Why this matters: [implication]
**Convergent Insights (Different Paths, Same Destination):**
• [Shared conclusion]
- Via expertise domain 1: [reasoning]
- Via expertise domain 2: [different reasoning]
- Synthesis: [why both paths validate the conclusion]
**Productive Tensions:**
• [Specific disagreement]
- Position A: [claim + reasoning]
- Position B: [competing claim + reasoning]
- Nature of disagreement: [factual/values/definitional]
- Context sensitivity: [conditions favoring each position]
**Reframed Questions:**
• Original question: [initial framing]
• Better question: [reframing that emerged from discussion]
• Why this reframe matters: [what it reveals]
**Actionable Next Steps:**
• [Concrete investigation to resolve empirical question]
• [Specific stakeholder to consult for values question]
• [Targeted research to test competing hypothesis]
```
---
## 3. Synthesis Generation Algorithm
### Step 1: Extract Positions
```
For each expert response:
1. Identify main claims (thesis)
2. Note supporting evidence/reasoning
3. Extract caveats and boundary conditions
4. Capture confidence levels (explicit or implicit)
5. Note any questions raised
```
### Step 2: Map Agreement Space
> [Implementation Choice] The categorization of agreement types (Explicit, Implicit, Partial) and their tags are design patterns that facilitate synthesis generation, not research prescriptions.
**Explicit Agreement:**
- Same conclusion AND same reasoning
- Tag: `[STRONG_CONSENSUS]`
**Implicit Agreement:**
- Same conclusion BUT different reasoning paths
- Tag: `[CONVERGENT]`
- Action: Synthesize to show how different domains validate same point
**Partial Agreement:**
- Agreement on subset of claims
- Tag: `[PARTIAL]`
- Action: Clearly separate agreed vs. contested elements
### Step 3: Categorize Disagreements
> [Implementation Choice] The five-way categorization of disagreement types (Factual, Values-Based, Definitional, Methodological, Scope) is a design pattern. Research supports explicit deliberation and context-dependent conclusions, but doesn't prescribe these specific categories.
**Factual Disagreements:**
- Different empirical claims
- Different predictions
- Tag: `[EMPIRICAL]`
- Resolution path: Identify evidence that would settle the dispute
**Values-Based Disagreements:**
- Different optimization criteria
- Different stakeholder priorities
- Tag: `[VALUES]`
- Resolution path: Make trade-offs explicit, identify decision-makers
**Definitional Disagreements:**
- Different meanings for key terms
- Different problem framings
- Tag: `[DEFINITIONAL]`
- Resolution path: Standardize definitions or acknowledge multiple valid framings
**Methodological Disagreements:**
- Different approaches to solving same problem
- Tag: `[METHODOLOGICAL]`
- Resolution path: Identify contexts where each approach excels
**Scope Disagreements:**
- Different boundaries for the problem
- Tag: `[SCOPE]`
- Resolution path: Explicitly define scope or address multiple scopes
### Step 4: Extract Emergent Insights
Look for:
- **Hidden assumptions made explicit**: "Both positions assume X, which may not hold if..."
- **Reframings**: "This isn't really about A vs. B, it's about optimizing for C vs. D"
- **Novel synthesis**: "Combining Expert A's insight with Expert B's constraint suggests..."
- **Meta-insights**: "Our disagreement reveals that we're solving different problems"
- **Contextual boundaries**: "Position A holds in context X, Position B holds in context Y"
### Step 5: Formulate Open Questions
Good open questions:
- Point to specific unknowns
- Suggest concrete resolution paths
- Are necessary for making progress
- Cannot be answered with current information
Template:
```
"To resolve [tension], we need to determine [specific unknown].
This could be answered by [concrete action/research/consultation]."
```
### Step 6: Generate Synthesis Statement
**For Consensus:**
```
"All experts agree that [conclusion] because [synthesized reasoning].
This is significant because [implication for the original question]."
```
**For Majority View:**
```
"Experts A and B argue [position] based on [reasoning],
while Expert C contends [alternative] based on [different reasoning].
The majority view is more persuasive in this case because [justification],
though Expert C's caveat about [concern] remains valid."
```
**For Unresolved Tension:**
```
"This panel reveals a fundamental trade-off between [value A] and [value B].
In contexts prioritizing [A], approach X is superior because [reasoning].
In contexts prioritizing [B], approach Y is superior because [reasoning].
Organizations should choose based on [specific decision criteria]."
```
---
## 4. Final Consensus Generation
### Consensus Labeling
> [Implementation Choice] The five consensus labels below (UNANIMOUS, STRONG, MAJORITY, CONTESTED, CONTEXT-DEPENDENT) are design patterns that operationalize research findings about argument-quality weighting and context-dependent conclusions. Adjust labels as needed.
Use precise labels:
**UNANIMOUS CONSENSUS:**
- All experts agree on conclusion and reasoning
- High confidence in actionability
- Format: "All N experts agree..."
**STRONG CONSENSUS:**
- All experts agree on conclusion
- Different reasoning paths converge
- Format: "Unanimous conclusion via multiple reasoning paths..."
**MAJORITY VIEW:** `[Research-Backed: CISC confidence weighting, but avoid majority suppression per Wu et al.]`
- Most experts agree (e.g., 3 of 4)
- Minority view acknowledged
- Format: "The majority position (N of M experts) is... However, Expert X notes..."
**CONTESTED:**
- Fundamental disagreement remains
- May be irreducible due to values/context
- Format: "Experts are split between Position A and Position B, reflecting a genuine trade-off..."
**CONTEXT-DEPENDENT:** `[Research-Backed: acknowledging "it depends" with conditions per research]`
- Different answers for different contexts
- Not a failure to reach consensus, but recognition of complexity
- Format: "The optimal approach depends on [specific factors]..."
### Weighted Summary Template
```markdown
## Final Synthesis
**Question:** [Original question clearly restated]
**Consensus Level:** [LABEL from above]
**Core Conclusion:**
[1-2 sentence answer that directly addresses the question]
**Supporting Analysis:**
From [Expertise Domain A]:
• [Key insight specific to this domain]
• [Reasoning and evidence]
From [Expertise Domain B]:
• [Key insight specific to this domain]
• [How it complements or tensions with Domain A]
**Integrated Perspective:**
[2-3 paragraphs synthesizing the above into coherent answer]
**Actionable Recommendations:**
1. [Concrete action item with reasoning]
2. [Concrete action item with conditions]
3. [Concrete action item with success criteria]
**Key Caveats:**
• [Boundary condition that would change the recommendation]
• [Assumption that, if false, would invalidate conclusion]
**Unresolved Questions:**
• [What we still don't know]
• [How to find out]
**Confidence Assessment:**
• High confidence in: [specific aspects]
• Moderate confidence in: [specific aspects]
• Low confidence in: [specific aspects]
• Requires further research: [specific aspects]
```
### Expertise-Weighted Synthesis
> [Research-Backed] Weighting expert opinions by domain relevance and argument quality is supported by CISC (ACL 2025) research on confidence weighting. This counteracts the conformity bias identified in multi-agent systems.
When expert relevance varies by topic:
```markdown
**On [Specific Sub-Question A]:**
Primary expertise: [Expert X's domain]
Conclusion: [Answer weighted toward Expert X's view]
Supporting view: [How other experts' perspectives complement this]
**On [Specific Sub-Question B]:**
Primary expertise: [Expert Y's domain]
Conclusion: [Answer weighted toward Expert Y's view]
Conflict note: [If Expert X disagrees, explain why Y's expertise dominates here]
```
---
## 5. Handling Unresolved Tensions
### When to Agree to Disagree
Legitimate reasons for persistent disagreement:
1. **Genuine Trade-offs**: Not every problem has one right answer
2. **Different Stakeholders**: Different constituencies have different needs
3. **Uncertainty**: Insufficient evidence to resolve empirical question
4. **Values Pluralism**: Multiple valid ethical frameworks
**Template:**
```markdown
**Unresolved Tension: [Topic]**
This panel reveals a genuine [trade-off/values conflict/empirical uncertainty].
Position A: [summary]
- Optimizes for: [value/goal]
- Strongest in context: [conditions]
- Risk if chosen: [downside]
Position B: [summary]
- Optimizes for: [different value/goal]
- Strongest in context: [different conditions]
- Risk if chosen: [downside]
**Decision Framework:**
Choose Position A if your context includes: [specific criteria]
Choose Position B if your context includes: [specific criteria]
**Red Flags:**
Avoid Position A if: [warning sign]
Avoid Position B if: [warning sign]
```
### Constructive "It Depends" Conclusions
> [Research-Backed] Acknowledging context-dependency rather than forcing binary consensus improves accuracy according to multi-agent debate research. The key is specifying WHAT it depends on.
> [Implementation Choice] The decision matrix format below is a design pattern for operationalizing context-dependent conclusions, not a research prescription.
Bad: "It depends on your situation."
Good: "The optimal approach depends on three specific factors:"
**Template:**
```markdown
**Context-Dependent Recommendation**
The answer to [question] depends on:
**Factor 1: [Specific, measurable attribute]**
• If [condition A]: Choose approach X because [reasoning]
• If [condition B]: Choose approach Y because [reasoning]
**Factor 2: [Specific, measurable attribute]**
• If [condition C]: This reinforces X because [reasoning]
• If [condition D]: This reinforces Y because [reasoning]
**Decision Matrix:**
| Factor 1 | Factor 2 | Recommendation | Confidence |
|----------|----------|----------------|------------|
| A | C | Strong X | High |
| A | D | Lean X | Medium |
| B | C | Lean Y | Medium |
| B | D | Strong Y | High |
**How to Determine Your Context:**
1. [Specific diagnostic question for Factor 1]
2. [Specific diagnostic question for Factor 2]
```
### Framing Trade-offs (Not False Dichotomies)
**Trade-off Template:**
```markdown
**Trade-off: [Value A] vs. [Value B]**
This is not a choice between right and wrong, but between two legitimate goals.
**Maximizing [Value A]:**
- Approach: [how to do this]
- Gains: [what you get]
- Costs: [what you sacrifice from Value B]
- Best for: [organizational context]
**Maximizing [Value B]:**
- Approach: [how to do this]
- Gains: [what you get]
- Costs: [what you sacrifice from Value A]
- Best for: [different organizational context]
**Balanced Approach:**
- [How to get 80% of both, if possible]
- [Or explain why this trade-off is zero-sum]
**Warning:** [Explain why trying to fully optimize both simultaneously will fail]
```
---
## 6. Templates for Synthesis Statements
### Agreement Templates
**Strong Consensus:**
```
"All experts converge on [conclusion]. [Expert A] reaches this through [domain A reasoning], while [Expert B] arrives via [domain B reasoning]. This convergence across methodologies strengthens confidence in [conclusion]."
```
**Qualified Agreement:**
```
"Experts agree on [core principle] but differ on [implementation details]. The consensus is that [general approach] is sound, with the caveat that [important boundary condition]."
```
### Disagreement Templates
**Productive Tension:**
```
"Expert A argues [position] based on [reasoning], while Expert B contends [alternative] based on [different reasoning]. Rather than a contradiction, this reveals [deeper insight about the problem space]."
```
**Irreducible Conflict:**
```
"This panel exposes a fundamental tension between [framework A] and [framework B]. Organizations must choose based on whether they prioritize [criterion from A] or [criterion from B]."
```
### Synthesis Templates
**Integrative Synthesis:**
```
"Combining Expert A's insight that [X] with Expert B's observation that [Y] suggests a synthesis: [Z]. This approach preserves the strengths of both positions while avoiding their respective weaknesses."
```
**Contextual Synthesis:**
```
"The apparent disagreement between Expert A and Expert B resolves when we recognize they're addressing different contexts. In [context 1], Expert A's approach of [X] is optimal because [reason]. In [context 2], Expert B's approach of [Y] is optimal because [different reason]."
```
**Sequential Synthesis:**
```
"Experts A and B aren't contradicting each other but describing sequential stages. Expert A's recommendation applies during [phase/condition 1], while Expert B's recommendation becomes relevant at [phase/condition 2]."
```
### Insight Emergence Templates
**Hidden Assumption Revealed:**
```
"The expert dialogue reveals a critical assumption: [assumption]. Both positions assume [X], but if [X] doesn't hold, then [implication for recommendations]."
```
**Reframing:**
```
"Initially framed as '[original question],' the expert discussion reveals the real question is '[better framing].' This reframe matters because [why this changes the answer]."
```
**Novel Integration:**
```
"Neither expert proposed [approach Z] individually, but Expert A's point about [X] combined with Expert B's point about [Y] suggests [Z] as a superior synthesis that [benefits]."
```
---
## 7. Examples: Raw Discussion → Synthesis
### Example 1: Software Architecture Decision
**Raw Discussion:**
**Expert A (Security Specialist):**
"Use a monolithic architecture. Microservices multiply your attack surface and make it harder to maintain consistent security policies. Each service is a potential breach point."
**Expert B (Scalability Engineer):**
"Microservices are essential for scale. Monoliths become bottlenecks and single points of failure. You need independent scaling of components."
**Expert C (Cost Optimizer):**
"Start with a well-structured monolith. Microservices have significant operational overhead—orchestration, monitoring, service mesh. Only justify the cost when monolith constraints actually manifest."
**Synthesis:**
```markdown
📋 Round 1 Synthesis:
**Agreements:**
• All experts agree that architecture should match organizational capabilities and actual requirements, not theoretical preferences
• Unanimous that premature architectural complexity is harmful
**Productive Tensions:**
• Security vs. Scalability: Expert A prioritizes attack surface minimization (favoring monolith), Expert B prioritizes independent scaling (favoring microservices)
• Timing disagreement: When to make the transition (if ever)
**Open Questions:**
• What is the specific scale threshold that justifies microservices complexity in this domain?
• What security controls make microservices acceptable from Expert A's perspective?
**Emerging Insights:**
• The disagreement is less "monolith vs. microservices" and more about WHEN and WHETHER transition is justified
• Expert C's "well-structured monolith" bridges the gap—modularity can be achieved within monolith, enabling later extraction if needed
• Security and scalability aren't inherently opposed; they're optimization criteria with different architectural implications
**Follow-up for Round 2:**
• Expert A: What specific security controls would make microservices acceptable?
• Expert B: At what user/transaction volume do monolith bottlenecks become real rather than theoretical?
• Expert C: Can you outline a "modular monolith" approach that preserves migration optionality?
```
**Final Consensus (After Round 2):**
```markdown
## Final Synthesis: Architecture Decision
**Consensus Level:** STRONG CONSENSUS (with implementation variations)
**Core Conclusion:**
Start with a modular monolith and transition to microservices only when specific, measurable constraints appear, not based on anticipated scale.
**Integrated Perspective:**
All experts converged on a "modular monolith first" approach after Round 2 dialogue. The architecture should feature:
1. **Clear internal boundaries** (Expert C's contribution): Structure the monolith as independently deployable modules from day one
2. **Security-first design** (Expert A's contribution): Implement security controls that work in both monolith and microservices contexts—zero trust networking, service-to-service authentication
3. **Observability infrastructure** (Expert B's contribution): Instrument for performance bottleneck detection so you know WHEN to split
**Transition Triggers** (Consensus from Round 2):
Migrate to microservices when you observe:
- Specific service scaling independently of others (measured via resource utilization)
- Teams blocked waiting for other teams' code freezes/releases (measured via deployment frequency)
- Monolith deployment time exceeding 30 minutes despite optimization
- Database connection pool exhaustion despite vertical scaling
**Actionable Recommendations:**
1. **Immediate:** Structure monolith with clear module boundaries, separate databases per domain
2. **Month 1-3:** Implement observability to detect transition triggers
3. **Ongoing:** Review metrics quarterly against transition triggers
4. **If triggers fire:** Extract highest-value service first (usually the scaling bottleneck)
**Key Caveats:**
• Assumes team size < 50 developers; larger teams may justify microservices earlier
• Assumes brownfield migration cost is acceptable; greenfield with distributed team might start distributed
• Requires discipline to maintain module boundaries in monolith
**Confidence Assessment:**
• High confidence: Modular monolith is correct starting point for most teams
• High confidence: Listed triggers are valid transition signals
• Moderate confidence: Specific numeric thresholds (30 min deploy, etc.) are context-dependent
• Requires case-by-case: Exact timeline for transition
```
---
### Example 2: Ethical Dilemma (Values-Based Disagreement)
**Raw Discussion:**
**Expert A (Privacy Advocate):**
"Absolutely do not implement user tracking. Privacy is a fundamental right. The fact that competitors do it doesn't make it ethical. Use privacy-preserving analytics methods."
**Expert B (Business Strategist):**
"Without understanding user behavior, you can't optimize the product. Users benefit from personalization. The question isn't whether to track, but how to do it transparently with informed consent."
**Synthesis:**
```markdown
📋 Round 1 Synthesis:
**Agreements:**
• Both experts agree transparency is essential (Expert A: as part of minimization; Expert B: as part of consent)
• Both agree "what competitors do" is not an ethical justification
**Productive Tensions:**
• Fundamental values conflict: Expert A prioritizes privacy as non-negotiable right; Expert B sees it as one value among others (including user experience)
• Different definitions of "harm": Expert A sees tracking itself as harm; Expert B sees only non-consensual or opaque tracking as harmful
**Open Questions:**
• Are there privacy-preserving analytics methods that satisfy both experts?
• What level of granularity in consent satisfies Expert A's concerns?
**Emerging Insights:**
• This is a genuine values conflict, not resolvable through technical means alone
• The answer may depend on: (a) legal jurisdiction, (b) user demographics, (c) sensitivity of domain
• Both experts might accept differential privacy or federated learning approaches
**Reframed Question:**
Original: "Should we implement user tracking?"
Better: "What analytics are essential for product improvement, and what's the minimum viable data collection to achieve that?"
**Follow-up for Round 2:**
• Expert A: What analytics approaches would you find acceptable?
• Expert B: What's the minimum data needed for your optimization goals?
• Both: Can differential privacy or on-device processing satisfy both criteria?
```
**Final Consensus:**
```markdown
## Final Synthesis: User Analytics Decision
**Consensus Level:** CONTEXT-DEPENDENT (irreducible values tension)
**Core Conclusion:**
The panel reveals a legitimate trade-off between privacy and personalization. Organizations must choose based on their domain, legal context, and user expectations.
**Supporting Analysis:**
From Privacy Perspective (Expert A):
• Tracking carries inherent risks: data breaches, de-anonymization, surveillance capitalism
• Users cannot meaningfully consent to complex data processing
• Privacy-preserving alternatives exist and should be default
From Business Perspective (Expert B):
• Product improvement requires understanding user behavior
• Transparent, consensual tracking can benefit users through personalization
• Absolute privacy prohibition may harm competitiveness and user experience
**Integrated Framework:**
Both experts agreed in Round 2 on this decision framework:
**Tier 1 (Always Acceptable):**
- Aggregated, anonymous analytics (no user identifiers)
- On-device analytics that never leave user's system
- Differential privacy with strong epsilon guarantees (ε < 1)
**Tier 2 (Acceptable with Strong Safeguards):**
- Pseudonymized analytics with informed, granular consent
- Minimal data collection (only what's necessary for stated purpose)
- User-accessible data deletion and export
- External privacy audits
**Tier 3 (Unacceptable to Expert A, Acceptable to Expert B):**
- Cross-site tracking
- Selling user data to third parties
- Opaque data practices
**Decision Criteria:**
**Choose Tier 1 if:**
- You're in healthcare, finance, or other sensitive domains
- Your users are privacy-conscious (measured via surveys)
- You're subject to GDPR, CCPA, or similar regulations with strict interpretation
**Choose Tier 2 if:**
- You need personalization for core product value
- You have resources for robust consent management
- Your domain is competitive and users expect personalization
**Avoid Tier 3:**
- Both experts agree these practices are ethically questionable
- Legal risk is high and increasing
**Key Insight:**
The experts' disagreement isn't about facts but about whether privacy is a trumping value or one value among others. Your organization's position on this philosophical question should guide the decision.
**Unresolved Tension:**
Expert A maintains that even "consensual" tracking in Tier 2 is problematic due to power imbalances and dark patterns. Expert B maintains that blanket prohibition in Tier 1 paternalistically denies users the choice to trade privacy for features. This tension is irreducible.
**Actionable Recommendation:**
1. Start with Tier 1 approaches
2. Measure whether Tier 1 analytics provide sufficient insight for product decisions
3. Only implement Tier 2 if you can demonstrate Tier 1 is insufficient AND you have resources for robust safeguards
4. Submit Tier 2 implementation to external privacy audit before deployment
```
---
## 8. Quality Checklist for Syntheses
Before finalizing a synthesis, verify:
- [ ] **Actionable**: Does it provide specific next steps, not just observations?
- [ ] **Preserves Nuance**: Are important caveats and boundary conditions included?
- [ ] **Not Wishy-Washy**: If it says "it depends," does it specify on what and how to decide?
- [ ] **Acknowledges Uncertainty**: Are confidence levels appropriate?
- [ ] **Shows Emergence**: Does it contain insights not present in individual expert statements?
- [ ] **Respects Expertise**: Are domain-specific insights weighted appropriately?
- [ ] **Identifies Open Questions**: Are unresolved issues clearly stated with resolution paths?
- [ ] **Avoids False Balance**: If experts genuinely disagree, is that acknowledged rather than papered over?
- [ ] **Contextually Grounded**: Are recommendations tied to specific conditions/contexts?
- [ ] **Directly Answers Question**: Does the synthesis actually address what was asked?
---
## 9. Anti-Patterns and Failure Modes
### Synthesis Red Flags
**Red Flag: Vague Integration**
```
❌ BAD: "Both experts make valid points. The truth is somewhere in between."
✅ GOOD: "Expert A's point holds in [context X] while Expert B's point holds in [context Y].
Specifically: when [measurable condition], choose A; when [different condition], choose B."
```
**Red Flag: Ignoring Disagreement**
```
❌ BAD: "Experts agree on the general approach." [when they actually fundamentally disagree]
✅ GOOD: "Experts agree on [specific point] but fundamentally disagree on [specific other point].
This disagreement stems from [values/factual/definitional difference]."
```
**Red Flag: Premature Consensus**
```
❌ BAD: "After one round, the panel reaches consensus that [complex question] should be handled by [simple answer]."
✅ GOOD: "Round 1 identifies key tensions: [list]. Round 2 explores [specific tension].
Consensus emerges after [specific insight] resolves the apparent contradiction."
```
**Red Flag: Abstraction Without Grounding**
```
❌ BAD: "The synthesis is to adopt a holistic, integrated approach that balances all stakeholder needs."
✅ GOOD: "Stakeholder A needs [specific thing], stakeholder B needs [different thing].
Satisfy A by [concrete action], satisfy B by [different concrete action].
Where they conflict (e.g., on [specific issue]), prioritize A if [condition], B if [other condition]."
```
**Red Flag: Expertise Mismatch**
```
❌ BAD: Treating all expert opinions equally when question clearly falls in one expert's domain
✅ GOOD: "On this [domain-specific question], Expert X's perspective is most relevant because [reasoning].
Expert Y's perspective, while valuable for [different aspect], is less applicable here because [reasoning]."
```
---
## 10. Meta-Synthesis: When the Process Itself Provides Insight
Sometimes the PROCESS of attempting synthesis reveals something important:
**Pattern: Persistent Disagreement → Values Discovery**
```markdown
"The panel's inability to reach consensus on [question] is itself informative.
It reveals that [question] is not a technical question but a values question.
The disagreement reflects legitimate differences in [value A vs. value B]."
```
**Pattern: Easy Consensus → Deeper Questions**
```markdown
"The panel quickly agreed on [answer], which suggests this is well-understood territory.
However, this easy consensus raises a more interesting question: [deeper question that experts hadn't considered]."
```
**Pattern: Confusion → Ambiguity Discovery**
```markdown
"Experts initially talked past each other because [term X] means [definition A] in Expert A's domain
but [definition B] in Expert B's domain. Disambiguating this term reveals we're actually asking two
separate questions: [question 1] and [question 2]."
```
**Pattern: Expertise Gaps**
```markdown
"All experts acknowledged this question requires input from [missing expertise domain].
The panel can address [subset of question] but [other aspect] requires [specific additional expert]."
```
---
## Conclusion
Effective synthesis is an active, creative process, not passive summarization. It requires:
1. **Precise categorization** of agreements, disagreements, and their natures
2. **Emergence detection** to find insights beyond what any expert stated
3. **Context sensitivity** to recognize when different answers apply in different situations
4. **Intellectual honesty** to acknowledge genuine uncertainties and trade-offs
5. **Actionability** to provide specific guidance, not abstract observations
The Hegelian dialectic succeeds when synthesis genuinely transcends thesis and antithesis—when the expert panel collectively reaches understanding that no individual expert possessed at the start.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |