bayesian-synthesis

Bayesian Synthesis

The Arbiter persona uses explicit Bayesian reasoning to synthesize multiple perspectives into a calibrated verdict. Use this approach when you need to make decisions under uncertainty.

The Bayesian Method

1. State your prior: What do you believe before seeing any arguments? (0-100%)
2. Update on evidence: For each new piece of evidence, adjust your belief
3. State your posterior: Final belief after all updates, with confidence interval
4. Identify remaining uncertainty: What would change your mind?

When to Use

• Decision-making under uncertainty: Multiple valid options, unclear trade-offs
• Forecasting: Predicting outcomes with limited information
• Evidence synthesis: Combining multiple studies or data sources
• Belief calibration: Checking if your confidence matches available evidence

How to Use

Full Council (recommended for important decisions)

council_query(
    question="What is the probability that Ethereum's market cap surpasses Bitcoin's by 2028?",
    context="Consider technical fundamentals, adoption metrics, regulatory landscape, and historical precedent."
)

The Arbiter's synthesis will include:

• PRIOR: Starting belief before arguments
• EVIDENCE UPDATES: How each persona's argument shifts the belief
• POSTERIOR: Final calibrated estimate
• KEY DISAGREEMENTS: Unresolved tensions
• FINAL VERDICT: Actionable recommendation

Quick Evaluation

council_evaluate(
    content="[your analysis here]",
    question="Is this analysis well-calibrated?",
    criteria=["calibration", "evidence_quality", "uncertainty_handling"]
)

Claim Review With Evidence

council_review_claim(
    claim="This library supports JSON mode on every OpenAI-compatible provider",
    context="We plan to rely on response_format for production parsing",
    evidence_search=true
)

Use verified_sources for retrieved evidence. Treat URLs in sources as model-cited unless they also appear in verified_sources.