Architectural Differences Between Full, Quick, and Duo Deliberation Modes in the Council of High Intelligence

The Council of High Intelligence implements three distinct deliberation architectures—Full, Quick, and Duo—each defined in SKILL.md with unique panel sizes, round structures, anonymization rules, and consensus mechanisms tailored to different problem complexities.

The 0xNyk/council-of-high-intelligence repository provides a multi-agent deliberation system where the architectural differences between full, quick, and duo deliberation modes determine how reasoning panels are constructed, how rounds execute, and how final consensus is reached. These modes are codified in the core specification file SKILL.md and supported by the 18 persona definitions in the agents/ directory and provider configurations in configs/auto-route-defaults.yaml.

Panel Composition and Selection Logic

Full Mode (--full)

In Full Mode, the system assembles either all 18 council members or a specified triad/explicit list during STEP 0 before any analysis begins. This mode supports a domain-weight seat that receives 1.5× weight during voting, creating a non-uniform influence distribution for specialized expertise.

Quick Mode (--quick)

Quick Mode defaults to the best-matching triad—for example, the architecture triad—without explicit user selection. Unlike Full Mode, Quick Mode applies uniform weighting with no domain-weight seat privilege, streamlining the selection process for time-sensitive queries.

Duo Mode (--duo)

Duo Mode collapses the panel to exactly two members selected either via the --members flag or algorithmically matched against the Duo Polarity Pairs table defined in lines 12‑24 of the Duo specification section. This binary pairing is designed for dialectic opposition rather than broad consensus.

Round Structure and Execution Flow

Full Mode Execution

Full Mode executes three distinct rounds:

1. Round 1: Independent analysis from all panel members
2. Round 2: Cross-examination with anonymization (members see only Member A/B labels)
3. Round 3: Final crystallization and position statements

The complete execution flow spans STEP 0 through STEP 9, including problem parsing, provider routing, restatement, enforcement checks, and weighted tie-breaking.

Quick Mode Execution

Quick Mode compresses the process into two rounds:

1. Round 1: Rapid analysis that embeds a brief problem restatement
2. Round 2: Final positions delivered with anonymization

The chairman synthesis occurs after a weighted tally, following QUICK STEP 3 and subsequent voting logic.

Duo Mode Execution

Duo Mode maintains three rounds but changes their nature:

1. Round 1: Opening positions from both participants
2. Round 2: Direct response where each member receives the opponent's full Round 1 output and must defend, concede, or refine their position
3. Round 3: Final statements

This structure eliminates anonymization to preserve the dialectic's authenticity.

Anonymization and Anti-Conformity Mechanisms

Full Mode applies anonymization in Round 2 to prevent conformity bias, requiring each member to critique at least two peers while viewing only Member A/B labels rather than identities. Quick Mode similarly applies anonymization in Round 2 (the "final positions" step) for the same anti-conformity directive.

Duo Mode explicitly rejects anonymization; with only two participants, each knows the opponent's identity, which is essential for a true dialectic exchange rather than blind peer review.

Consensus Mechanisms and Chairman Integration

Weighted Voting in Full and Quick Modes

Both Full and Quick modes implement weighted voting after the final round. Each member emits a STANCE: line, and the system computes a weighted tally in STEP 6 (1.0 per member, 1.5 for the domain-weight seat). Consensus requires ≥ 2/3 of total weight to resolve ties or determine majority positions.

Dialectic Synthesis in Duo Mode

Duo Mode abandons weighted tallies entirely. The two STANCE: lines are combined through chairman synthesis, but with a hard constraint: the chairman must not be either duo participant. The final verdict emerges from the dialectic exchange rather than numeric calculation.

Chairman Selection

Chairman selection occurs in STEP 1.7 across all modes, utilizing auto-selection, override flags, or configuration-driven logic from configs/auto-route-defaults.yaml. In Duo Mode, this selection includes the additional constraint that the chairman cannot be one of the two active participants.

Practical Implementation Examples

The following commands demonstrate the architectural differences between full, quick, and duo deliberation modes in practice:

# Full mode – explicit triad selection with deep analysis

/council --triad architecture "Should we adopt a monorepo for our codebase?"

# Quick mode – default triad for rapid, time-sensitive answers

/council --quick "Should we add caching to the auth flow?"

# Duo mode – polarity pair selection for binary debate

/council --duo "Should we rewrite the monolith as micro-services?"

Summary

• Full Mode (--full): 18 members or triad, three rounds with Round 2 anonymization, weighted voting (1.5× domain seat), and STEP 0‑9 execution for complex, high-stakes problems.
• Quick Mode (--quick): Triad default, two rounds with embedded restatement, anonymized final positions, and streamlined STEP 6 voting for time-sensitive queries.
• Duo Mode (--duo): Two members via Polarity Pairs (lines 12‑24), three rounds of direct dialectic without anonymization, chairman exclusion constraint, and synthesis-based verdicts for binary decisions.

Frequently Asked Questions

What is the main difference between Full and Quick mode?

The primary distinction lies in round depth and panel flexibility. Full Mode executes three rounds with explicit cross-examination and supports 18 members or custom triads with weighted domain expertise, while Quick Mode compresses the process into two rounds using a default triad and embedded restatement for faster turnaround.

Why does Duo mode not use anonymization?

Duo Mode intentionally disables anonymization because the architecture depends on a direct dialectic between two opposing perspectives. According to the specification in SKILL.md, knowing the opponent's identity is essential for members to defend, concede, or refine positions against specific arguments rather than anonymous critiques.

How does the chairman selection differ across modes?

All modes use STEP 1.7 for chairman selection, but Duo Mode adds a unique constraint: the chairman cannot be either of the two participating members. This ensures an impartial synthesis of the dialectic, whereas Full and Quick modes only prohibit the chairman from panel participation generally.

Can I mix panel selection methods between modes?

No, each mode enforces specific selection logic. Full Mode accepts --triad or explicit lists processed during STEP 0, Quick Mode auto-selects the best-matching triad, and Duo Mode requires exactly two members via --members or the Duo Polarity Pairs table algorithm.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}