Features – Deliberation modes index

8 Deliberation Modes

Each mode implements a distinct debate protocol optimized for specific decision types.

Comparison Table

---

Quick

Single-round rapid analysis. One model, one API call. Bypasses challenge, rebuttal, voting, and convergence entirely. Use for simple factual queries and time-constrained decisions.

Council (Default)

Multi-round deliberation with full 8-phase pipeline: PROPOSAL → CHALLENGE → REBUTTAL → EVALUATION → VOTING → AGGREGATION → CONVERGENCE → OUTPUT. Condorcet voting with Ranked Pairs fallback. Convergence: 0.4 × ranking + 0.35 × proposal + 0.25 × concession ≥ 0.85. The standard mode for most decisions.

Deep

Same phases as Council but 5 rounds instead of 3, plus sub-agent research capability. Convergence may trigger early. Use for security audits, compliance reviews, mission-critical decisions.

Blind

Anonymous evaluation — proposals stripped of model identity before judging. Judge evaluates in multiple orderings to prevent anchoring bias. model_id → anonymous_1, anonymous_2. Use for fair comparison, legal review, unbiased evaluation.

Red Team

Adversarial testing with 8 attack categories: LOGICAL_FLAW, EDGE_CASE, SECURITY_VULN, BIAS_DETECTION, HALLUCINATION_PROBE, PROMPT_INJECTION, ROBUSTNESS_TEST, CONSISTENCY_CHECK. Attackers probe, defenders respond, judges evaluate. Use for security assessment and code review.

Jury

Mandatory dissent reporting via agglomerative clustering. Must declare majority AND minority positions. No "unanimous" unless mathematically verified (cluster threshold ≥ 0.5). Used by risk assessment and finance templates.

Market (Truth Market)

Prediction market mechanism. Models assign probability distributions, update via log-opinion pooling. Converges when max(|P_t - P_{t-1}|) < 5%. Produces probability distribution rather than single winner. Use for forecasting and probability estimation.

Auto

Intelligent routing via complexity scoring. Features: token count, code presence, stakes keywords (medical/legal/financial/security). Score < 0.3 → Quick/1 model. Score 0.3-0.6 → Council/3 models. Score ≥ 0.6 → Deep/3-5 models.