board-meeting（董事会会议）

Board Meeting Protocol

Structured multi-agent deliberation that prevents groupthink, captures minority views, and produces clean, actionable decisions. Every phase has a purpose, a format, and rules that cannot be skipped.

Keywords

board meeting, executive deliberation, strategic decision, C-suite, multi-agent, founder review, decision extraction, independent perspectives, groupthink prevention, synthesis, critic analysis, structured deliberation

The 6-Phase Protocol

PHASE 1: Context Gathering
    |
PHASE 2: Independent Contributions (ISOLATED)
    |
PHASE 3: Critic Analysis (Executive Mentor)
    |
PHASE 4: Synthesis (Chief of Staff)
    |
PHASE 5: Founder Review (FULL STOP -- human decides)
    |
PHASE 6: Decision Extraction and Logging

Phase 1: Context Gathering

Purpose: Load all relevant context before anyone contributes.

Step 1: Load company context (if exists)
Step 2: Load decision history (Layer 2 ONLY -- NEVER raw transcripts)
Step 3: Reset session state -- no bleed from previous conversations
Step 4: Present agenda and activated roles
Step 5: Wait for founder confirmation before proceeding

Role Activation Matrix

Not all roles attend every meeting. Select based on topic:

Maximum attendees: 6 roles per meeting. More than 6 creates noise, not insight.

Phase 2: Independent Contributions (ISOLATED)

Critical Rule: No cross-pollination. Each advisor contributes without seeing others' outputs. This is the primary groupthink prevention mechanism.

Contribution Order

1. Research/data gathering (if needed)
2. CMO  -- market perspective
3. CFO  -- financial perspective
4. CEO  -- strategic perspective
5. CTO  -- technical perspective
6. COO  -- operational perspective
7. CHRO -- people perspective
8. CRO  -- revenue perspective
9. CISO -- security/risk perspective
10. CPO -- product perspective

Contribution Format (Strict)

Each advisor's contribution must follow this exact format:

## [ROLE] -- [DATE]

Key Points (maximum 5):
  1. [Finding] -- Confidence: [High/Medium/Low] -- Source: [data source]
  2. [Finding] -- Confidence: [High/Medium/Low] -- Source: [data source]
  3. [Finding] -- Confidence: [High/Medium/Low] -- Source: [data source]

Recommendation: [Clear position statement]
Confidence: [High / Medium / Low]
Key Assumption: [The one assumption this recommendation depends on]
What Would Change My Mind: [Specific condition or data point]

Reasoning Techniques by Role

Phase 3: Critic Analysis

Purpose: The Executive Mentor receives ALL Phase 2 outputs simultaneously and performs adversarial review.

Critic Checklist

Critic Output Format

## CRITIC ANALYSIS

Consensus Assessment:
  [Genuine agreement / Suspicious alignment / Split decision]

Unvalidated Assumptions:
  1. [Assumption shared by multiple advisors but not verified]
  2. [Assumption]

Missing Perspectives:
  - [Voice or data point not represented]

Unmentioned Risks:
  - [Risk nobody raised]

Domain Violations:
  - [If any agent operated outside their domain]

The Uncomfortable Truth:
  [The one thing nobody wants to say but needs to be said]

Phase 4: Synthesis

Purpose: Chief of Staff combines all inputs into a decision-ready format.

Synthesis Structure

## BOARD MEETING SYNTHESIS
Topic: [topic]
Date: [date]
Attendees: [roles]

### Decision Required
[One sentence: what must be decided]

### Perspectives Summary
| Role | Position | Confidence | Key Concern |
|------|----------|-----------|-------------|
| [Role] | [1-line summary] | [H/M/L] | [Top concern] |
| [Role] | [1-line summary] | [H/M/L] | [Top concern] |

### Where They Agree
[2-3 consensus points]

### Where They Disagree
[Named conflicts with each side's reasoning]
[What the disagreement is really about]

### Critic's View
[The uncomfortable truth from Phase 3]

### Recommended Decision
[Clear recommendation with rationale]

### Action Items (if approved)
1. [Action] -- Owner: [role] -- Deadline: [date]
2. [Action] -- Owner: [role] -- Deadline: [date]
3. [Action] -- Owner: [role] -- Deadline: [date]

### Your Call
[If you disagree with the recommendation, here are alternatives:]
Option A: [description] -- Trade-off: [what you gain/lose]
Option B: [description] -- Trade-off: [what you gain/lose]

Phase 5: Founder Review

FULL STOP. Wait for the founder. No agent acts beyond this point.

FOUNDER REVIEW

[Paste synthesis above]

Options:
  [A] Approve as recommended
  [M] Modify (specify changes)
  [R] Reject (specify reason)
  [Q] Ask follow-up question to specific role
  [D] Defer decision (specify timeline)

Phase 5 Rules

Phase 6: Decision Extraction

Purpose: After founder approval, extract and log all decisions.

Step 1: Write full transcript to Layer 1
  --> memory/board-meetings/YYYY-MM-DD-raw.md

Step 2: Run conflict detection against existing decisions
  --> Check for DO_NOT_RESURFACE violations
  --> Check for topic contradictions
  --> Check for owner conflicts

Step 3: Surface any conflicts to founder for resolution

Step 4: Append approved decisions to Layer 2
  --> memory/board-meetings/decisions.md

Step 5: Mark rejected proposals with DO_NOT_RESURFACE

Step 6: Confirm to founder:
  "Meeting concluded. Logged: [N] decisions, [M] action items,
   [K] DO_NOT_RESURFACE flags."

Failure Mode Reference

Meeting Cadence

Red Flags

• Board meetings consistently produce no decisions -- meeting is theater
• Same topic discussed in 3+ meetings -- decision avoidance
• Phase 2 contributions all align perfectly -- isolation was breached or topic is trivial
• No Phase 3 (critic) conducted -- groupthink risk
• Founder skipping Phase 5 -- decisions without accountability
• Decisions logged but never reviewed -- decision logger not functioning
• Meeting attendees always include all roles -- topic selection not working

Output Artifacts

Tool Reference

meeting_simulator.py

Validates role activation, contribution completeness, and phase sequencing.

# Simulate with defaults
python scripts/meeting_simulator.py

# Specify topic and complexity
python scripts/meeting_simulator.py --topic "Series B timing" --type fundraising --complexity 9

# Specify activated roles
python scripts/meeting_simulator.py --type m_and_a --roles CEO CFO CTO CHRO

# List all topic types and required roles
python scripts/meeting_simulator.py --list-topics

# JSON output
python scripts/meeting_simulator.py --type strategy --json

decision_tracker.py

Tracks board decisions, detects conflicts, flags overdue reviews and actions.

# Track demo decisions
python scripts/decision_tracker.py

# From decision log file
python scripts/decision_tracker.py --input decisions.json

# JSON output
python scripts/decision_tracker.py --json

complexity_scorer.py

Scores decision complexity to determine single/dual/multi-advisor or board routing.

# Score with CLI flags
python scripts/complexity_scorer.py --topic "Market expansion" --domains 2 --reversibility 2 --financial 1 --team 2 --urgency 0

# Add modifiers
python scripts/complexity_scorer.py --topic "Acquisition" --domains 2 --reversibility 2 --financial 2 --team 2 --urgency 1 --modifiers cross_functional external_stakeholders sets_precedent

# JSON output
python scripts/complexity_scorer.py --topic "Pricing change" --json

Troubleshooting

Success Criteria

• Every board meeting produces at least 1 logged decision with owner, deadline, and review date
• Phase 2 contributions are independently generated (zero cross-pollination incidents per quarter)
• Phase 3 critic analysis identifies at least 1 unvalidated assumption per meeting
• Founder approval/modification/rejection captured within 30 minutes of synthesis presentation
• Decision history has zero conflicting active decisions (conflicts detected and resolved)
• Meeting duration stays within 2 hours for standard strategic reviews, 1 hour for resolution meetings
• 90%+ of logged decisions have action items completed by their stated deadlines

Scope & Limitations

In Scope: Multi-agent deliberation protocol, role activation matrix, contribution formats, critic analysis, synthesis, decision extraction, decision conflict detection, meeting simulation.

Out of Scope: Actual AI agent orchestration (this is a protocol specification, not runtime code), real-time meeting facilitation, video/audio recording, external board member management.

Limitations: The protocol assumes all advisor contributions are available in text format. Complexity scoring provides routing guidance but cannot account for political dynamics. Decision conflict detection works on exact topic matching -- semantic conflicts across different topics require human judgment.

Integration Points