Roadmap: Enterprise Simulation & ML Ground Truth

This roadmap documents the completed feature waves and outlines the direction for future development.

Completed Features

v0.1.0 — Core Generation

• Statistical distributions: Benford’s Law compliance, log-normal mixtures, copulas
• Industry presets: Manufacturing, Retail, Financial Services, Healthcare, Technology
• Chart of Accounts: Small (~100), Medium (~400), Large (~2500) complexity levels
• Temporal patterns: Month-end/quarter-end volume spikes, business day calendars
• Master data: Vendors, customers, materials, fixed assets, employees
• Document flows: P2P (6 PO types, three-way match) and O2C (9 SO types, 6 delivery types, 7 invoice types)
• Intercompany: IC matching, transfer pricing, consolidation elimination entries
• Subledgers: AR (aging, dunning), AP (scheduling, discounts), FA (6 depreciation methods), Inventory (22 movement types, 4 valuation methods)
• Currency & FX: Ornstein-Uhlenbeck exchange rates, ASC 830 translation, CTA
• Period close: Monthly close engine, accruals, depreciation runs, year-end closing
• Balance coherence: Opening balances, running balance tracking, trial balance per period
• Anomaly injection: 60+ fraud types, error patterns, process issues with full labeling
• Data quality: Missing values (MCAR/MAR/MNAR), format variations, typos, duplicates
• Graph export: PyTorch Geometric, Neo4j, DGL with train/val/test splits
• Internal controls: COSO 2013 framework, SoD rules, 12 transaction + 6 entity controls
• Resource guards: Memory, disk, CPU monitoring with graceful degradation
• REST/gRPC/WebSocket server with authentication and rate limiting
• Desktop UI: Tauri/SvelteKit with configuration pages
• Python wrapper: Programmatic access with blueprints and config validation

v0.2.0 — Privacy & Standards

• Fingerprint extraction: Statistical properties from real data into .dsf files
• Differential privacy: Laplace and Gaussian mechanisms with configurable epsilon
• K-anonymity: Suppression of rare categorical values
• Fidelity evaluation: KS, Wasserstein, Benford MAD metric comparison
• Gaussian copula synthesis: Preserve multivariate correlations
• Accounting standards: Revenue recognition (ASC 606/IFRS 15), Leases (ASC 842/IFRS 16), Fair Value (ASC 820/IFRS 13), Impairment (ASC 360/IAS 36)
• Audit standards: ISA compliance (34 standards), analytical procedures, confirmations, opinions, PCAOB mappings
• SOX compliance: Section 302/404 assessments, deficiency matrix, material weakness classification
• Streaming output: CSV, JSON, NDJSON, Parquet streaming sinks with backpressure
• ERP output formats: SAP S/4HANA (BKPF, BSEG, ACDOCA, LFA1, KNA1, MARA), Oracle EBS (GL_JE_HEADERS/LINES), NetSuite

v0.3.0 — Fraud & Industry

• ACFE-aligned fraud taxonomy: Asset misappropriation, corruption, financial statement fraud calibrated to ACFE statistics
• Collusion modeling: 8 ring types, 6 conspirator roles, defection/escalation dynamics
• Management override: Fraud triangle modeling (pressure, opportunity, rationalization)
• Red flag generation: 40+ probabilistic indicators with Bayesian probabilities
• Industry-specific generators: Manufacturing (BOM, WIP, production orders), Retail (POS, shrinkage, loyalty), Healthcare (ICD-10, CPT, DRG, payer mix)
• Industry benchmarks: Pre-configured ML benchmarks per industry
• Banking/KYC/AML: Customer personas, KYC profiles, fraud typologies (structuring, funnel, layering, mule, round-tripping)
• Process mining: OCEL 2.0 event logs with P2P and O2C processes
• Evaluation framework: Auto-tuning with configuration recommendations from metric gaps
• Vendor networks: Tiered supply chains, quality scores, clusters
• Customer segmentation: Value segments, lifecycle stages, network positions
• Cross-process links: Entity graph, relationship strength, cross-process integration

v0.5.0 — AI & Advanced Features

• LLM-augmented generation: Pluggable provider abstraction for realistic vendor names, descriptions, and anomaly explanations
• Natural language configuration: Generate YAML configs from descriptions
• Diffusion model backend: Statistical diffusion with configurable noise schedules
• Hybrid generation: Blend rule-based and diffusion outputs
• Causal generation: Structural Causal Models, do-calculus interventions, counterfactual generation
• Federated fingerprinting: Secure aggregation for distributed data sources
• Synthetic data certificates: Cryptographic proof of DP guarantees with HMAC-SHA256
• Privacy-utility Pareto frontier: Automated exploration of optimal epsilon values
• Ecosystem integrations: Airflow, dbt, MLflow, Spark pipeline integration

v0.6.0–v0.8.x — Enterprise Process Chains & Localization

• Source-to-Contract (S2C): Spend analysis, sourcing projects, supplier qualification, RFx, bids, evaluation, contracts, catalogs, scorecards
• Hire-to-Retire (H2R): Payroll runs, time & attendance, expense reports, benefit enrollment
• Manufacturing: Production orders, BOM explosion, routing operations, WIP costing, quality inspections, cycle counting
• Universal OCPM: 12 process families with 101+ activities and 65+ object types
• Country packs: Pluggable JSON architecture with 10 built-in packs (US, DE, GB, FR, JP, CN, IN, IT, ES, CA)
• French GAAP (PCG): Plan Comptable General 2024 with FEC export
• German GAAP (HGB): SKR04 chart of accounts, Degressiv depreciation, GWG, GoBD export
• Generalized FrameworkAccounts: ~45 semantic accounts per framework

v0.9.0 — Performance & Quality

• ~2x single-threaded throughput via cached temporal CDF, fast Decimal, SmallVec, parallel generation
• ParallelGenerator trait with deterministic seed splitting
• RustGraph property mapping for 51 entity types, 28 relationship types with edge constraints
• Comprehensive edge-case hardening across all crates
• VAT line splitting, multipayment behavior, account-class fingerprinting

v0.10.0–v0.11.0 — Scenarios & Streaming

• Counterfactual simulation engine: Causal DAG with 17 financial nodes, 8 transfer functions, paired baseline/counterfactual generation
• Scenario CLI: datasynth-data scenario {list, validate, generate, diff}
• GenerationSession: Multi-period generation with checkpoint files and incremental append
• Fraud scenario packs: 5 built-in packs with deep-merge configuration
• StreamPipeline: Phase-aware streaming via PhaseSink trait
• OCEL 2.0 enrichment: Lifecycle state machines, correlation events, resource pool modeling

v1.0.0 — Release

• Process evolution & organizational events: Acquisitions, divestitures, mergers, reorganizations
• Disruption events: Outage, migration, process change, recovery, regulatory disruption
• Collusion ring generation: Coordinated fraud networks with escalation dynamics
• Bi-temporal vendor versioning: Valid-time/transaction-time dimension version chains
• Entity relationship graph: Strength scores and cross-process links
• Industry transaction factory: Industry-specific GL accounts per vertical
• Red flag indicators: Risk indicators on P2P/O2C document chains
• Counterfactual pairs: (original, mutated) journal entry pairs for ML training
• Performance optimizations, numeric safety hardening, code quality improvements
• Complete crate metadata and documentation overhaul

v1.4.0 — Realism & Coherence

• Cost center hierarchy generator: Parent/child trees with department mappings and GL assignments
• Employee change history: Title changes, salary adjustments, department transfers
• Multi-period balance carry-forward: Trial balance closing → opening balance propagation
• Dunning generator wiring: Dunning runs and letters after AR aging
• AR/AP reconciliation validation: Subledger totals vs GL control accounts
• Contract→PO linkage: Procurement contracts carry PO IDs for S2P chain traversal
• Moving-average inventory cost: AVCO updated on each goods receipt
• Production order ↔ inventory cross-refs: Bidirectional traceability
• ISA mappings output: 34 ISA standard reference records
• SoD/COSO control mappings: Automated export of conflict pairs and COSO mappings
• Graph export enhancements: JE→Employee (POSTED_BY), Control→JE (CONTROL_APPLIED) edges
• Consolidated financial statements: Standalone + consolidated with elimination schedules

v1.5.0 — Audit FSM Engine & Optimizer

• YAML-driven audit FSM engine (datasynth-audit-fsm): Loads ISA and IIA-GIAS methodology blueprints as event-sourced finite state machines
  • Financial Statement Audit (FSA): 9 procedures, 3 phases, 24 steps → 51 events, 1,916 artifacts
  • Internal Audit (IA): 34 procedures, 9 phases, 82 steps → 368 events, 1,891 artifacts
  • StepDispatcher: 135 command mappings to 14 pre-initialized audit generators
  • 8-state C2CE (Condition-Criteria-Cause-Effect) lifecycle for finding development
  • Self-loop handling with configurable max iterations
  • Continuous phase support (parallel execution for ethics, governance, quality)
  • Discriminator-based procedure filtering (categories, risk ratings, engagement types)
  • Generation overlay presets: default, thorough, rushed
  • Flat JSON event trail + OCEL 2.0 projection exports
  • Custom YAML blueprint support
• Audit FSM optimizer (datasynth-audit-optimizer): Graph analysis and Monte Carlo simulation
  • Blueprint → petgraph directed graph conversion
  • Shortest path analysis (BFS per procedure): FSA 27 min transitions, IA 101
  • Constraint-based path optimization with transitive precondition expansion
  • Monte Carlo simulation: bottleneck detection, revision hotspots, happy path identification
• Orchestrator integration: FSM engine wired into enhanced orchestrator with full artifact pipeline
• Blueprint repository: SyntheticDataBlueprints

Process Coverage (v2.0.0)

Cross-Process Integration

Strategic Roadmap

Waves 1 through 5 are complete as of v2.0.0. Wave 6 represents the research frontier.

v1.6.0 — Wave 1: Consolidation

Production-ready FSM pipeline.

End-to-End CLI Integration

• Full pipeline verification: Run datasynth-data generate with audit.fsm.enabled: true and verify the complete output directory (all 50+ artifact types written to audit/, event trail to audit/fsm_event_trail.json)
• Missing sink registrations: Ensure all ArtifactBag types flow through the standard output writer (CSV, JSON, Parquet)
• Config validation: Validate audit.fsm config section during datasynth-data validate
• Demo mode: Add FSM-enabled preset to --demo for instant evaluation

Blueprint Validation Tooling

• CLI command: datasynth-data audit validate-blueprint --file my_methodology.yaml — runs loader validation, reports cross-reference errors, DAG cycles, unreachable states
• Blueprint info: datasynth-data audit info --blueprint builtin:fsa — prints procedure count, phase structure, step commands, standards coverage
• Diff tool: Compare two blueprints to show added/removed procedures, changed transitions

IA Artifact Fidelity

• Richer IA command dispatch: Map IA-specific commands to specialized generators:
  • assess_universe_risks → risk universe document with entity-level risk ratings
  • develop_recommendations → structured recommendation artifacts with management response fields
  • draft_ia_charter → formal IA charter document with mandate, scope, authority
  • develop_annual_plan → audit plan artifact with resource allocation and timeline
• IA-specific workpaper sections: Extend WorkpaperSection with IA variants (Universe, Planning, Fieldwork, Monitoring, QA)
• Finding quality: C2CE findings should carry quantified financial impact, root cause categorization, and management action plan timelines

Graph Integration Depth

• Audit-specific graph edges: Step→Evidence (PRODUCED_BY), Finding→Risk (IDENTIFIED_FROM), Opinion→Finding (BASED_ON), Workpaper→Standard (COMPLIES_WITH)
• Temporal audit graph: Time-ordered engagement events as a temporal knowledge graph with TGN-compatible export
• Hypergraph audit nodes: Register all FSM artifact types (engagement, materiality, finding, opinion) as first-class hypergraph nodes

v1.7.0 — Wave 2: Audit Planning Optimization

The optimizer crate as a planning tool.

Resource-Constrained Optimization

• Cost model: Assign hour costs per procedure (partner hours, manager hours, staff hours) from overlay or blueprint metadata
• Staff availability constraints: Model team capacity (e.g., partner available 20h/week, 3 seniors available) and find feasible audit plans
• Budget optimization: Given a total hour budget, find the audit plan that maximizes risk coverage
• Critical path analysis: Identify the longest dependency chain that determines minimum engagement duration

Risk-Based Audit Scoping

• Risk-weighted procedure selection: Given a risk profile (industry, entity size, prior findings), use discriminators and Monte Carlo to recommend which procedures to include
• Coverage analysis: For a given scope, compute ISA/IIA-GIAS standards coverage percentage and identify uncovered requirements
• What-if analysis: “What happens to coverage if we drop procedure X?” — instant impact assessment via graph analysis

Multi-Engagement Portfolio

• Portfolio simulation: Generate N engagements with correlated parameters (same industry → correlated risk profiles, shared systemic findings)
• Resource pooling: Model shared audit team across engagements with scheduling constraints
• Portfolio risk heatmap: Aggregate risk across engagements to identify systemic exposure

v1.8.0 — Wave 3: Process Mining & Benchmarks

OCEL 2.0 projection for process mining research.

Reference Audit Event Logs

• Benchmark dataset generation: Produce standardized OCEL 2.0 audit event logs with known anomalies (skipped approvals, late postings, out-of-sequence steps) at configurable injection rates
• Conformance checking ground truth: Given a blueprint (normative model) and generated event log (with deviations), provide labeled conformance violations for evaluating process mining tools
• Multiple complexity levels: Simple (FSA, default overlay, no anomalies), Medium (FSA, rushed overlay, moderate anomalies), Complex (IA, mixed overlays, high anomaly rate)

Process Mining Tool Integration

• PM4Py native export: Direct export to PM4Py DataFrame format with object-centric support
• Celonis IBC format: Export compatible with Celonis Intelligent Business Cloud import
• ProM/XES export: Traditional single-object event log in XES format for backward compatibility
• Disco/Minit CSV: Flat case-activity-timestamp CSV for commercial process mining tools

Conformance Metrics

• Fitness score: Percentage of event traces that conform to the blueprint FSM
• Precision score: How much behavior the model allows beyond what was observed
• Generalization: Model behavior on unseen engagement configurations
• Anomaly detection benchmark: F1/precision/recall for process mining anomaly detectors against known injected anomalies

v1.9.0 — Wave 4: Learned & Adaptive Generation

Deterministic FSM framework combined with learned components.

Learned Overlay Parameters

• Engagement profile fitting: Given real audit engagement metadata (duration, findings count, revision frequency, team size), fit overlay parameters to reproduce those characteristics
• Industry-calibrated overlays: Pre-fitted overlays for financial services audits, manufacturing audits, technology audits based on aggregate engagement statistics
• Temporal drift: Overlay parameters that evolve over time (e.g., increasing regulatory scrutiny → more revision loops, longer durations)

LLM-Augmented Artifact Content

• Contextual narrative generation: Plug an LLM into the StepDispatcher to generate finding descriptions, management responses, workpaper narratives, and engagement letter prose
• ISA-grounded prompts: Each step’s standards references become prompt context, ensuring generated text cites the correct ISA paragraphs
• Deterministic fallback: LLM output is optional; the system produces valid artifacts with template text when LLM is unavailable
• Quality control: Generated narratives validated against step evidence requirements (finding must reference the evidence it was derived from)

Bidirectional Blueprint Discovery

• Process discovery from event logs: Given real audit event logs (anonymized), infer the underlying methodology blueprint (states, transitions, procedures)
• Blueprint comparison: Diff a discovered blueprint against a reference (ISA, IIA-GIAS) to identify deviations from standard methodology
• Methodology conformance scoring: Quantify how closely an organization’s actual audit practice matches the declared methodology

Adaptive Anomaly Calibration

• Reinforcement learning: Tune anomaly injection parameters such that downstream detector performance matches a target difficulty curve
• Curriculum generation: Progressive difficulty datasets — start with obvious anomalies, gradually increase subtlety
• Adversarial generation: Generate anomalies specifically designed to evade a given detector, for robustness testing

v2.0.0 — Wave 5: Enterprise Platform

Platform-level capabilities, now completed.

Continuous Audit Simulation

• Streaming engagement execution: FSM engine emits events via callback or mpsc channel, enabling WebSocket/Kafka/dashboard integration
• Live anomaly injection: inject_live_anomalies() marks existing events with anomaly flags after generation, simulating emerging risks
• Dashboard integration: Event stream compatible with Grafana, Splunk, or custom audit dashboards via the streaming API

Multi-Engagement Correlation

• Group audit coordination: ISA 600 group audit (group_audit module) with component auditors, each running their own FSM engagement, consolidated at group level
• Year-over-year engagement chains: yoy_chain module simulates sequential engagements with configurable finding carry-forward rates and trend tracking
• Portfolio simulation: portfolio module generates N engagements with correlated parameters and resource pooling
• Benchmark comparison: benchmark_comparison module runs side-by-side blueprint comparisons

Custom Blueprint Ecosystem

• 9 built-in blueprints: FSA, IA, KPMG, PwC, Deloitte, EY GAM Lite, SOC 2 Type II, PCAOB Integrated, Regulatory Exam
• Blueprint testing framework: test_blueprint() and test_all_builtins() in datasynth-audit-optimizer validate blueprints against expected metrics
• Blueprint repository: Community-contributed blueprints at SyntheticDataBlueprints

Additional Audit Frameworks

• PCAOB AS blueprint: US public company integrated audit with PCAOB-specific procedures and critical audit matters
• SOC 2 Type II blueprint: Service organization audit with AICPA Trust Services Criteria across 5 categories
• Regulatory exam blueprint: Banking supervision examination (OCC, Fed, FDIC, EBA) with CAMELS rating and BSA/AML review
• Big 4 methodology blueprints: KPMG, PwC, Deloitte, and EY GAM Lite ISA-aligned blueprints (non-proprietary structures)
• Analytics inventories: Data requirement and analytical procedure mappings for FSA, IA, SOC 2, PCAOB, and Regulatory frameworks

Wave 6: Research Frontier (v3.x)

Long-term research directions.

Knowledge Graph Completion Benchmarks

• Reference audit knowledge graphs: Fully provenanced graphs where every node, edge, and property traces to a known generative process
• Standardized KGC benchmarks: Evaluate knowledge graph construction and completion algorithms against known ground truth
• Multi-layer evaluation: Separate accuracy metrics for structural (entity/relationship), statistical (amount distributions), and normative (standards compliance) layers

Temporal Graph Networks for Audit

• TGN-compatible export: Audit event streams formatted for temporal graph network models
• Dynamic audit risk prediction: Train TGN models to predict engagement outcomes from partial event trails
• Early warning detection: Identify engagements heading toward adverse opinions from early-phase events

Federated Audit Simulation

• Cross-organization generation: Multiple organizations with shared auditor (simulates Big 4 portfolio)
• Privacy-preserving aggregation: Federated learning on audit engagement statistics without exposing individual client data
• Industry-level reference graphs: Aggregate knowledge graphs spanning multiple generated enterprises for sector-level analysis

Causal Audit Analytics

• Causal DAG for audit outcomes: Structural causal model linking engagement parameters (scope, team, timeline) to outcomes (findings, opinion type, duration)
• do-calculus interventions: “What would happen to the opinion if we doubled the substantive testing scope?”
• Counterfactual engagement pairs: Generate paired engagements (baseline vs intervention) for causal inference research

Implementation Status

Guiding Principles

• Enterprise realism: Simulate multi-entity, multi-region, multi-currency operations with coherent process flows
• ML ground truth: Capture true labels and causal factors for supervised learning, explainability, and evaluation
• Scalability: Handle large volumes with stable performance and reproducible results
• Backward compatibility: New features are additive; existing configs continue to work

Contributing

We welcome contributions to any area. See Contributing Guidelines for details.

To propose new features:

1. Open a GitHub issue with the enhancement label
2. Describe the use case and expected behavior
3. Reference relevant roadmap items if applicable

Feedback

Priorities are influenced by user feedback. Please share your use cases and requirements:

• GitHub Issues: Feature requests and bug reports
• GitHub: Issues

See Also

• Process Chains — Process chain architecture and coverage matrix
• S2P Spec — Source-to-Contract specification
• Process Chain Gaps — Detailed gap analysis