This section details on the original issue you should resolve

<issue_title>🤖 AI-Driven Multi-Iteration Political Analysis Engine for News Generation</issue_title>
<issue_description>## 📋 Issue Type
Feature / Architecture Enhancement

🎯 Objective

Redesign the news article generation pipeline to use AI-driven multi-iteration analysis, replacing static template-based content with dynamically generated political intelligence. All document analysis must use AI reasoning across multiple iterations to produce deeper, more nuanced political insights from every stakeholder perspective.

📊 Current State

• scripts/generate-news-enhanced/generators.ts (1,690 lines) uses single-pass content generation
• Deep-inspection articles use generateDeepPolicyAnalysis() for a single 600-char analysis per document
• SWOT entries are populated from document metadata with hardcoded fallback strings (SWOT_DEFAULTS)
• Strategic implications use template strings (en/sv/de/fr/es only) rather than AI reasoning
• Key takeaways are mechanical bullet lists from document type counts
• No iterative refinement — first-pass output is the final output

🚀 Desired State

Multi-Iteration AI Analysis Pipeline

1. First Pass — Data Collection & Classification: Fetch documents, classify by type/domain, detect policy areas
2. Second Pass — AI Deep Analysis: Use MCP AI tools to generate substantive political analysis for each document, including:
   • Legislative impact assessment
   • Cross-party implications
   • Historical context and precedent analysis
   • International comparison (EU, Nordic)
3. Third Pass — Cross-Document Synthesis: AI-driven synthesis across all documents to identify:
   • Policy convergence/divergence patterns
   • Coalition stress indicators
   • Emerging legislative trends
   • Stakeholder power dynamics
4. Fourth Pass — Quality Assurance & Refinement: AI reviews its own output for:
   • Factual accuracy against source documents
   • Balance across political perspectives
   • Analytical depth and insight quality
   • Consistent tone and style across languages

Architecture Changes

• Add scripts/generate-news-enhanced/ai-analysis-pipeline.ts — orchestrates multi-iteration AI analysis
• Enhance generateDeepInspectionContent() to call AI iteratively with context accumulation
• Replace SWOT_DEFAULTS hardcoded strings with AI-generated context-aware defaults
• Add scripts/generate-news-enhanced/analysis-cache.ts — cache intermediate AI analysis results
• Support --iterations=N CLI parameter (default: 3) for controlling analysis depth

AI Analysis Prompts Architecture

interface AIAnalysisIteration {
  iteration: number;
  context: string;          // Accumulated context from previous iterations
  documents: RawDocument[];  
  focusTopic: string | null;
  previousAnalysis?: string; // Output from previous iteration
  stakeholderPerspective: 'government' | 'opposition' | 'civil-society' | 'media' | 'international';
}

📊 CIA Data Integration Context

CIA Product(s): All 19 visualization products for comprehensive political intelligence
Data Source: json-export-specs/schemas/*.json
Methodology: Multi-perspective OSINT analysis per DATA_ANALYSIS_INTOP_OSINT.md

🔧 Implementation Approach

1. Create AIAnalysisPipeline class with configurable iteration count
2. Define AI prompt templates for each analysis phase
3. Implement context accumulation between iterations
4. Add quality scoring for AI outputs (reject below threshold)
5. Support all 14 languages with AI-native generation (not translation)
6. Add telemetry for iteration performance tracking
7. Maintain backward compatibility with existing article types

🤖 Recommended Agent

code-quality-engineer — Complex refactoring of the generation pipeline architecture

✅ Acceptance Criteria

• Multi-iteration pipeline generates deeper analysis than single-pass
• AI analysis produces unique, context-aware SWOT entries (no hardcoded fallbacks)
• Strategic implications are substantive political analysis, not template strings
• Quality scoring rejects shallow analysis and triggers re-analysis
• All 14 languages produce native-quality political analysis
• --iterations CLI parameter controls analysis depth
• Performance: ≤ 120s per article with 3 iterations
• All existing tests continue to pass
• TypeScript strict mode compliance

📚 References

• Repository: https://github.com/Hack23/riksdagsmonitor
• Current generators: scripts/generate-news-enhanced/generators.ts
• SWOT defaults: lines 1069-1086 in generators.ts
• CIA OSINT: https://github.com/Hack23/cia/blob/master/DATA_ANALYSIS_INTOP_OSINT.md
• ISMS Policy: https://github.com/Hack23/ISMS-PUBLIC/blob/main/Secure_Development_Policy.md

🏷️ Labels

type:feature, priority:high, component:data-integration</issue_description>

<agent_instructions>Implement multi-iteration AI analysis pipeline for ...