All files / scripts/ai-analysis pipeline.ts

82.82% Statements 82/99
73.8% Branches 31/42
83.33% Functions 10/12
83.69% Lines 77/92

Press n or j to go to the next uncovered block, b, p or k for the previous block.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279                                                                                                                  94x   94x 94x 314x 94x     94x 94x 94x 94x 94x 94x   94x                                                 13x 13x 13x   13x 12x                   1x     1x         1x     1x 1x 1x     1x 1x   1x     1x 1x                       1x                     11x 11x 11x   11x 33x 33x       33x       33x       33x       33x 66x 33x 9x 9x       11x 3x 3x     11x 2x 2x     11x 11x 9x 9x 2x         11x 5x 5x     11x     11x       11x 11x                                         18x                             13x   13x 13x 13x   13x 10x 10x 10x     13x 13x 7x 7x 7x 7x     13x    
/**
 * @module ai-analysis/pipeline
 * @description Thin orchestrator for the AI-first analysis pipeline.
 *
 * Implements the `AnalysisPipeline` interface by delegating to focused
 * bounded-context modules:
 *   - `swot/`          — stakeholder SWOT construction, enrichment, confidence
 *   - `domains/`       — policy assessment, watch points, EU/Nordic comparative
 *   - `visualisation/` — mindmap branches, dashboard data
 *
 * Lifecycle for a `deep` analysis:
 *   1. `analyzeDocuments`     — iteration 1: initial extraction
 *   2. `refineAnalysis`       — iteration 2: SWOT enrichment from full text
 *   3. `validateCompleteness` — iteration 3: quality validation
 *
 * @author Hack23 AB
 * @license Apache-2.0
 */
 
import type { RawDocument } from '../data-transformers/types.js';
import { detectPolicyDomains, detectNarrativeFrames } from '../data-transformers/policy-analysis.js';
 
import type {
  AnalysisPipeline,
  AnalysisPipelineOptions,
  AnalysisResult,
  ValidationResult,
} from './types.js';
 
import {
  buildStakeholderSwot,
  refineStakeholderSwot,
  calculateConfidenceScore,
} from './swot/index.js';
 
import { buildPolicyAssessment, buildWatchPoints } from './domains/index.js';
 
import {
  narrativeFramesLabel,
  buildMindmapBranches,
  buildDashboardData,
} from './visualisation/index.js';
 
import {
  docType,
  isMetadataEnriched,
  hasFullTextContent,
} from './helpers.js';
 
// ---------------------------------------------------------------------------
// Iteration 1: analyzeDocuments
// ---------------------------------------------------------------------------
 
async function analyzeDocuments(
  docs: RawDocument[],
  options: AnalysisPipelineOptions,
): Promise<AnalysisResult> {
  const { lang, focusTopic: topic } = options;
 
  const domains = (() => {
    const all = new Set<string>();
    docs.forEach(d => detectPolicyDomains(d, lang).forEach(dom => all.add(dom)));
    return [...all].slice(0, 8);
  })();
 
  const stakeholderSwot = buildStakeholderSwot(docs, topic, lang, domains);
  const policyAssessment = buildPolicyAssessment(docs, topic, lang, options.depth);
  const watchPoints = buildWatchPoints(docs, topic, lang);
  const mindmapBranches = buildMindmapBranches(docs, topic, policyAssessment.domains, lang);
  const dashboardData = buildDashboardData(docs, topic, lang);
  const confidenceScore = calculateConfidenceScore(docs, stakeholderSwot);
 
  return {
    stakeholderSwot,
    policyAssessment,
    mindmapBranches,
    dashboardData,
    watchPoints,
    confidenceScore,
    iterationsCompleted: 1,
    completedAt: new Date().toISOString(),
    lang,
    documentCount: docs.length,
    enrichedCount: docs.filter(isMetadataEnriched).length,
    focusTopic: topic,
  };
}
 
// ---------------------------------------------------------------------------
// Iteration 2: refineAnalysis
// ---------------------------------------------------------------------------
 
async function refineAnalysis(
  initial: AnalysisResult,
  docs: RawDocument[],
  options: AnalysisPipelineOptions,
): Promise<AnalysisResult> {
  const { lang, focusTopic: topic } = options;
  const fullTextDocs = docs.filter(hasFullTextContent);
  const metadataCount = docs.filter(isMetadataEnriched).length;
 
  if (fullTextDocs.length === 0) {
    return {
      ...initial,
      iterationsCompleted: 2,
      completedAt: new Date().toISOString(),
      enrichedCount: metadataCount,
      confidenceScore: calculateConfidenceScore(docs, initial.stakeholderSwot),
      policyAssessment: buildPolicyAssessment(docs, topic, lang, options.depth),
    };
  }
 
  const refined = { ...initial, iterationsCompleted: 2, completedAt: new Date().toISOString() };
 
  // Delegate SWOT enrichment to swot module
  refined.stakeholderSwot = refineStakeholderSwot(
    refined.stakeholderSwot, fullTextDocs, topic, lang,
  );
 
  // Rebuild watch points with enriched context
  refined.watchPoints = buildWatchPoints(docs, topic, lang);
 
  // Recalculate confidence and metadata
  refined.confidenceScore = calculateConfidenceScore(docs, refined.stakeholderSwot);
  refined.enrichedCount = metadataCount;
  refined.policyAssessment = buildPolicyAssessment(docs, topic, lang, options.depth);
 
  // Collect narrative frames for mindmap enrichment
  const allFrames = new Set<string>();
  fullTextDocs.slice(0, 20).forEach(d => detectNarrativeFrames(d).forEach(f => allFrames.add(f)));
 
  refined.mindmapBranches = buildMindmapBranches(docs, topic, refined.policyAssessment.domains, lang);
 
  // Add civil society branch if we have external docs
  const civilSocietyDocs = fullTextDocs.filter(d => ['ext', 'fpm', 'eu'].includes(docType(d)));
  Iif (civilSocietyDocs.length > 0 && allFrames.size > 0) {
    const hasBranch = refined.mindmapBranches.some(b => b.label === narrativeFramesLabel(lang));
    if (!hasBranch) {
      refined.mindmapBranches.push({
        label: narrativeFramesLabel(lang),
        color: 'orange',
        icon: '🎯',
        items: [...allFrames].slice(0, 6),
      });
    }
  }
 
  return refined;
}
 
// ---------------------------------------------------------------------------
// Iteration 3: validateCompleteness
// ---------------------------------------------------------------------------
 
async function validateCompleteness(
  analysis: AnalysisResult,
  docs: RawDocument[],
): Promise<ValidationResult> {
  const issues: string[] = [];
  const suggestions: string[] = [];
  let score = 100;
 
  for (const sh of analysis.stakeholderSwot) {
    const { strengths, weaknesses, opportunities, threats } = sh.swot;
    Iif (strengths.length === 0) {
      issues.push(`${sh.name}: no strengths entries`);
      score -= 5;
    }
    Iif (weaknesses.length === 0) {
      suggestions.push(`${sh.name}: consider adding weakness analysis`);
      score -= 2;
    }
    Iif (opportunities.length === 0) {
      suggestions.push(`${sh.name}: consider adding opportunity analysis`);
      score -= 2;
    }
    Iif (threats.length === 0) {
      suggestions.push(`${sh.name}: consider adding threat analysis`);
      score -= 2;
    }
    const allEntries = [...strengths, ...weaknesses, ...opportunities, ...threats];
    const allPlaceholders = allEntries.length > 0 && allEntries.every(e => e.sourceDocIds.length === 0);
    if (allPlaceholders) {
      issues.push(`${sh.name}: all SWOT entries lack document evidence (sourceDocIds empty) — enrich with document content`);
      score -= 10;
    }
  }
 
  if (analysis.policyAssessment.domains.length === 0) {
    issues.push('No policy domains detected — document titles may need more descriptive text');
    score -= 5;
  }
 
  if (analysis.watchPoints.length === 0) {
    suggestions.push('No watch points generated — check for actionable document types (prop, bet, sfs)');
    score -= 3;
  }
 
  const fullTextAvailable = docs.filter(hasFullTextContent).length;
  if (analysis.documentCount > 0 && analysis.enrichedCount === 0) {
    issues.push('No documents have been enriched — analysis quality is limited to raw metadata');
    score -= 15;
  } else if (Ianalysis.enrichedCount > 0 && fullTextAvailable === 0) {
    suggestions.push('Documents are metadata-enriched but lack full text — consider using include_full_text=true for deeper analysis');
    score -= 5;
  }
 
  if (analysis.confidenceScore < 30) {
    issues.push(`Low confidence score (${analysis.confidenceScore}/100) — fetch more documents or enrich with full text`);
    score -= 5;
  }
 
  Iif (analysis.enrichedCount >= 3) {
    score = Math.min(100, score + 5);
  }
  Iif (fullTextAvailable >= 2) {
    score = Math.min(100, score + 3);
  }
 
  const finalScore = Math.max(0, score);
  return {
    passed: finalScore >= 60,
    score: finalScore,
    issues,
    suggestions,
  };
}
 
// ---------------------------------------------------------------------------
// Exported pipeline singleton
// ---------------------------------------------------------------------------
 
/**
 * The default AI analysis pipeline implementation.
 *
 * Usage in generators:
 * ```ts
 * import { aiAnalysisPipeline } from '../ai-analysis/pipeline.js';
 * const result = await aiAnalysisPipeline.analyzeDocuments(docs, { depth, lang, focusTopic });
 * ```
 */
export const aiAnalysisPipeline: AnalysisPipeline = {
  analyzeDocuments,
  refineAnalysis,
  validateCompleteness,
};
 
/**
 * Run the full analysis pipeline according to the specified depth.
 * Returns the final AnalysisResult, optional ValidationResult, and
 * per-iteration timing data for audit metadata.
 */
export async function runAnalysisPipeline(
  docs: RawDocument[],
  options: AnalysisPipelineOptions,
): Promise<{ analysis: AnalysisResult; validation: ValidationResult | null; iterationDurationsMs: number[] }> {
  const iterationDurationsMs: number[] = [];
 
  const t1 = Date.now();
  let analysis = await aiAnalysisPipeline.analyzeDocuments(docs, options);
  iterationDurationsMs.push(Date.now() - t1);
 
  if (options.depth !== 'quick') {
    const t2 = Date.now();
    analysis = await aiAnalysisPipeline.refineAnalysis(analysis, docs, options);
    iterationDurationsMs.push(Date.now() - t2);
  }
 
  let validation: ValidationResult | null = null;
  if (options.depth === 'deep') {
    const t3 = Date.now();
    validation = await aiAnalysisPipeline.validateCompleteness(analysis, docs);
    iterationDurationsMs.push(Date.now() - t3);
    analysis = { ...analysis, iterationsCompleted: 3, completedAt: new Date().toISOString() };
  }
 
  return { analysis, validation, iterationDurationsMs };
}