1. BLOCK COMPLETION
   ├── User opens block → AI suggests content
   ├── Detect bias → Show warning if solution-first
   ├── Save → Extract assumptions
   └── Calculate score → Update progress

2. AI ASSISTANCE
   ├── Problem: Extract from interviews, website
   ├── Customer: Build psychographic from JTBD
   ├── UVP: Generate 3 options (≤120 chars)
   ├── Solution: Suggest MVP features
   └── All blocks: Industry-specific guidance

3. ASSUMPTION TRACKING
   ├── Extract implicit assumptions per block
   ├── Score by Impact × Uncertainty
   ├── Prioritize top 3 for testing
   └── Link to Validation Lab

const BLOCK_WEIGHTS = {
  problem: 0.20,           // Most critical
  customer_segments: 0.15,
  uvp: 0.15,
  solution: 0.10,
  channels: 0.10,
  revenue: 0.10,
  costs: 0.05,
  metrics: 0.10,
  unfair_advantage: 0.05
};

function calculateCanvasScore(canvas: LeanCanvas): number {
  return Object.entries(BLOCK_WEIGHTS).reduce((score, [block, weight]) => {
    const blockScore = evaluateBlockCompleteness(canvas[block]);
    return score + (blockScore * weight);
  }, 0);
}

const SOLUTION_FIRST_PATTERNS = [
  /we (will|are going to) build/i,
  /our (app|platform|software|tool)/i,
  /using (AI|ML|blockchain)/i,
  /features? (include|like)/i
];

function detectSolutionBias(problemBlock: string): boolean {
  return SOLUTION_FIRST_PATTERNS.some(p => p.test(problemBlock));
}

// Actions
- 'suggest_content': Generate block content suggestions
- 'extract_assumptions': Pull testable assumptions from block
- 'detect_bias': Check for solution-first thinking
- 'score_canvas': Calculate completeness score
- 'version_canvas': Save versioned snapshot

lean_canvas_versions (
  id uuid PRIMARY KEY,
  startup_id uuid REFERENCES startups,
  version int,
  canvas_data jsonb,
  score int,
  assumptions jsonb,
  created_at timestamptz
)