proffesor-for-testing__agentic-qe/docs/implementation/ddd/defect-intelligence.md

Defect Intelligence Domain

Bounded Context Overview

Domain: Defect Intelligence Responsibility: Defect prediction, root cause analysis, and regression risk assessment Location: src/domains/defect-intelligence/

The Defect Intelligence domain uses machine learning to predict defects, analyze root causes, assess regression risk, and cluster similar defects for pattern recognition.

Ubiquitous Language

Term	Definition
Defect Prediction	ML-based probability of bugs in code
Root Cause	Underlying reason for a defect
Regression Risk	Probability of breaking existing functionality
Defect Cluster	Group of related defects with common factors
Contributing Factor	Element that influenced a defect
Defect Pattern	Learned pattern for defect prevention
Impact Area	Code region affected by changes

Domain Model

Aggregates

PredictionResult (Aggregate Root)

Complete defect prediction analysis.

interface PredictionResult {
  predictions: FilePrediction[];
  modelConfidence: number;
  factors: string[];
  llmAnalysis?: LLMDefectAnalysis;    // ADR-051
}

RootCauseAnalysis (Aggregate Root)

Comprehensive root cause investigation.

interface RootCauseAnalysis {
  defectId: string;
  rootCause: string;
  confidence: number;
  contributingFactors: ContributingFactor[];
  relatedFiles: string[];
  recommendations: string[];
  timeline: TimelineEvent[];
}

Entities

FilePrediction

Defect prediction for a single file.

interface FilePrediction {
  file: string;
  probability: number;
  riskLevel: Severity;
  factors: { name: string; contribution: number }[];
  recommendations: string[];
}

DefectCluster

Group of related defects.

interface DefectCluster {
  id: string;
  label: string;
  defects: string[];
  commonFactors: string[];
  suggestedFix: string;
}

DefectPattern

Learned pattern for defect prevention.

interface DefectPattern {
  id: string;
  name: string;
  indicators: string[];
  frequency: number;
  prevention: string;
}

Value Objects

ContributingFactor

Immutable factor contributing to a defect.

interface ContributingFactor {
  readonly factor: string;
  readonly impact: 'high' | 'medium' | 'low';
  readonly evidence: string[];
}

RegressionRisk

Immutable regression risk assessment.

interface RegressionRisk {
  readonly overallRisk: number;
  readonly riskLevel: Severity;
  readonly impactedAreas: ImpactedArea[];
  readonly recommendedTests: string[];
  readonly confidence: number;
}

ImpactedArea

Code area affected by changes.

interface ImpactedArea {
  readonly area: string;
  readonly files: string[];
  readonly risk: number;
  readonly reason: string;
}

LLMDefectAnalysis

AI-enhanced defect analysis (ADR-051).

interface LLMDefectAnalysis {
  readonly riskFactors: string[];
  readonly reviewFocusAreas: string[];
  readonly similarHistoricalDefects: string[];
  readonly confidenceLevel: number;
  readonly explanation: string;
}

Domain Services

DefectIntelligenceAPI

Primary API for the domain.

interface DefectIntelligenceAPI {
  predictDefects(request: PredictRequest): Promise<Result<PredictionResult, Error>>;
  analyzeRootCause(request: RootCauseRequest): Promise<Result<RootCauseAnalysis, Error>>;
  analyzeRegressionRisk(request: RegressionRequest): Promise<Result<RegressionRisk, Error>>;
  clusterDefects(request: ClusterRequest): Promise<Result<DefectClusters, Error>>;
  learnPatterns(request: LearnRequest): Promise<Result<LearnedDefectPatterns, Error>>;
}

Domain Events

Event	Trigger	Payload
DefectPredictedEvent	Prediction complete	{ file, probability, riskLevel }
RootCauseIdentifiedEvent	Root cause found	{ defectId, rootCause, confidence }
RegressionRiskAssessedEvent	Risk calculated	{ changeset, riskLevel, impactedAreas }
DefectPatternLearnedEvent	New pattern learned	{ patternId, indicators, frequency }

Prediction Features

The defect prediction model uses these features:

Feature	Weight	Description
change_frequency	0.20	How often file changes
complexity	0.15	Cyclomatic complexity
code_churn	0.15	Lines added/removed recently
author_experience	0.10	Author's familiarity with file
dependency_count	0.10	Number of dependencies
test_coverage	0.10	Current test coverage
historical_defects	0.15	Past defects in file
code_age	0.05	Time since last major refactor

Clustering Methods

Method	Use Case	Algorithm
semantic	Text similarity	TF-IDF + K-Means
behavioral	Failure patterns	DBSCAN on execution traces
temporal	Time-based grouping	Hierarchical clustering

Context Integration

Upstream Dependencies

• Code Intelligence: Dependency graphs, complexity metrics
• Test Execution: Failure history
• Learning Optimization: Pattern storage

Downstream Consumers

• Test Generation: Targets high-risk areas
• Quality Assessment: Risk scores for gates

Anti-Corruption Layer

The domain abstracts different bug tracking systems (Jira, GitHub Issues, Linear) through the DefectInfo interface.

Task Handlers

Task Type	Handler	Description
predict-defects	predictDefects()	ML-based prediction
analyze-root-cause	analyzeRootCause()	Root cause analysis
assess-regression	analyzeRegressionRisk()	Regression risk
cluster-defects	clusterDefects()	Defect clustering
learn-patterns	learnPatterns()	Pattern learning

Risk Calculation

function calculateRegressionRisk(
  changeset: string[],
  dependencies: DependencyGraph,
  history: DefectHistory
): RegressionRisk {
  const impactedAreas: ImpactedArea[] = [];
  let totalRisk = 0;

  for (const file of changeset) {
    // Direct impact
    const directDeps = dependencies.getOutgoing(file);
    const historicalDefects = history.getDefectsFor(file);

    const fileRisk = calculateFileRisk({
      changeSize: getChangeSize(file),
      complexity: getComplexity(file),
      testCoverage: getCoverage(file),
      historicalDefectRate: historicalDefects.length / history.totalPeriod,
      dependentCount: directDeps.length,
    });

    totalRisk = Math.max(totalRisk, fileRisk);

    if (fileRisk > 0.3) {
      impactedAreas.push({
        area: file,
        files: [file, ...directDeps],
        risk: fileRisk,
        reason: determineRiskReason(fileRisk, historicalDefects),
      });
    }
  }

  return {
    overallRisk: totalRisk,
    riskLevel: mapRiskToSeverity(totalRisk),
    impactedAreas,
    recommendedTests: selectTestsForRisk(impactedAreas),
    confidence: calculateConfidence(changeset, history),
  };
}

ADR References

• ADR-051: LLM-powered root cause analysis
• ADR-006: HNSW indexing for pattern search