How raghavendrabhatbs/decision_rewind works
DECISION-REWIND is a cybersecurity research prototype that simulates correcting historical event data and selectively rewinding affected decisions. Users work through the React dashboard: generate or select an experiment, propose feature corrections, preview counterfactual decisions, execute a selective rewind, and inspect verification and audit results. FastAPI orchestrates dataset/model operations and the decision workflow; provenance reachability informs replay and recovery, while deterministic verification checks results. SQLite-backed stores preserve experiment state and logs. AI explanation is a separate, non-authoritative capability. Links reflect README workflow documentation and shown source references; unsampled implementation details are not inferred.
Main entry points: main.py HTTP API · graph.py Dependency graph · replay.py Counterfactual replay · App.tsx Research dashboard
Components 16
Dashboard
- Research dashboard
frontend/src/App.tsx
API orchestration
- HTTP API
backend/app/main.py - Request schemas
backend/app/schemas/models.py
Decision and recovery
- Event generator
backend/app/dataset/generator.py - Model registry
backend/app/ml/model_registry.py - Model workbench
backend/app/ml/workbench_model.py - Dependency graph
backend/app/provenance/graph.py - Counterfactual replay
backend/app/counterfactual/replay.py - Selective recovery
backend/app/recovery/recovery.py - Deterministic verifier
backend/app/verification/verifier.py
Data and assistance
- Experiment store
backend/app/services/dataset_store.py - Event store
backend/app/services/store.py - Universal audit log
backend/app/services/universal_log.py - AI explanation
backend/app/llm/provider.py
Other components
- Researcher
- Groq API
Connections 21
- Researcher → Research dashboard: uses
README.md - Research dashboard → HTTP API: submits workflow
README.md - HTTP API → Request schemas: validates requests
backend/app/main.py - HTTP API → Event generator: generates events
backend/app/main.py - HTTP API → Model workbench: trains models
backend/app/main.py - HTTP API → Model registry: uses registry
backend/app/main.py - HTTP API → Counterfactual replay: analyzes corrections
backend/app/main.py - HTTP API → Selective recovery: selects rewind
backend/app/main.py - HTTP API → Deterministic verifier: verifies recovery
backend/app/main.py - HTTP API → Dependency graph: builds graph
backend/app/main.py - HTTP API → Experiment store: persists experiments
backend/app/main.py - HTTP API → Event store: persists events
backend/app/main.py - HTTP API → Universal audit log: records activity
backend/app/main.py - HTTP API → AI explanation: requests explanations
backend/app/main.py - Counterfactual replay → Event generator: recomputes decisions
backend/app/counterfactual/replay.py - Counterfactual replay → Model registry: uses registry
backend/app/counterfactual/replay.py - Counterfactual replay → Dependency graph: checks reachability
backend/app/counterfactual/replay.py - Selective recovery → Dependency graph: checks reachability
backend/app/recovery/recovery.py - Deterministic verifier → Dependency graph: checks reachability
backend/app/verification/verifier.py - Experiment store → Dependency graph: stores dependencies
backend/app/services/dataset_store.py - AI explanation → Groq API: requests completion
backend/app/llm/provider.py