// Parallel execution of all checks
const checkResults = await Promise.allSettled([
  completenessCheck.run(entries, context),
  duplicateCheck.run(entries, context),
  categorizationCheck.run(entries, context),
  calculationCheck.run(entries, context),
  freshnessCheck.run(entries, context),
  anomalyCheck.run(entries, context),
]);

// Weighted score aggregation
const weights = {
  calculation: 2.0,   // Math errors are serious
  duplicate: 1.5,     // Duplicates affect totals
  anomaly: 1.2,       // Unusual patterns need attention
  completeness: 1.0,  // Standard importance
  categorization: 1.0,
  freshness: 0.8,     // Less critical
};

// Evaluation pipeline
async evaluate(decomposition: TaskDecomposition): EvaluationResult {
  // Fast heuristics first (~50ms)
  const heuristicIssues = this.runHeuristicChecks(decomposition);

  // LLM evaluation for deeper analysis (~500ms)
  const llmAssessment = await this.llmEvaluate(decomposition, {
    model: modelTiers.getModelId('reasoning'),
    temperature: 0.3,  // Consistent evaluation
  });

  // Merge results and determine action
  return this.mergeAssessments(heuristicIssues, llmAssessment);
}

// Confidence level determination
function determineConfidence(results: CheckResults): ConfidenceLevel {
  const errorCount = results.filter(r => r.severity === 'error').length;
  const warningCount = results.filter(r => r.severity === 'warning').length;
  const weightedScore = calculateWeightedScore(results);

  // Any errors = low confidence
  if (errorCount > 0) return 'low';

  // Score below threshold = low confidence
  if (weightedScore < 0.7) return 'low';

  // Many warnings or medium score = medium confidence
  if (warningCount > 3 || weightedScore < 0.9) return 'medium';

  // High score with few issues = high confidence
  return 'high';
}