SyntheticDataBlueprints

Curated FSM and YAML blueprint definitions for DataSynth — the high-performance Rust synthetic enterprise data generator.

This repository provides ready-to-use audit methodology blueprints, generation profiles, fraud scenarios, and causal DAGs for generating statistically realistic, fully interconnected financial data at scale.

Repository Structure

SyntheticDataBlueprints/
├── blueprints/
│   ├── fsm/                              # Finite State Machine definitions
│   │   ├── big4/                         # Big 4 ISA-complete firm-style blueprints
│   │   │   ├── kpmg_clara.yaml           # KPMG Clara (44 procs, 728 steps)
│   │   │   ├── pwc_aura.yaml            # PwC Aura/Halo (44 procs, 729 steps)
│   │   │   ├── deloitte_omnia.yaml      # Deloitte Omnia (46 procs, 733 steps)
│   │   │   └── ey_gam_lite.yaml         # EY GAM Lite (52 procs, 757 steps)
│   │   ├── external-audit/
│   │   │   ├── generic_fsa.yaml          # Generic ISA Financial Statement Audit
│   │   │   ├── overlay_default.yaml      # Default generation overlay
│   │   │   ├── overlay_thorough.yaml     # Thorough engagement overlay
│   │   │   └── overlay_rushed.yaml       # Rushed engagement overlay
│   │   └── internal-audit/
│   │       └── generic_ia.yaml           # IIA-GIAS Internal Audit (34 procs, 82 steps)
│   ├── generation/                       # DataSynth generation configs
│   │   ├── profiles/                     # Complete use-case profiles
│   │   ├── scenarios/                    # Industry & fraud scenario overlays
│   │   └── regional/                     # Country-specific naming data
│   └── causal/                           # Causal DAG definitions
├── explorer/                             # Interactive visualization UI
├── schemas/                              # Validation schemas (planned)
├── CHANGELOG.md
└── LICENSE                               # Apache 2.0

Big 4 Methodology Blueprints

ISA-complete audit methodology blueprints styled after each Big 4 firm's publicly documented platform and approach. Each blueprint contains 37 ISA procedures (one per standard, 702 requirement-level steps) plus firm-specific extra procedures reflecting documented platform capabilities.

Blueprint	Firm	Procedures	Steps	Events	Artifacts	Firm-Specific Extras
kpmg_clara.yaml	KPMG	44	728	891	30,371	Sentinel independence, MindBridge AI scoring, SoD analysis, forensic analytics, EQCR multi-point review, FRA disclosure
pwc_aura.yaml	PwC	44	729	976	37,055	FRISK 13-factor assessment, Halo journal/population/3-way/outlier analytics, QRP hot review, ECR
deloitte_omnia.yaml	Deloitte	46	733	958	34,236	Cortex data ingestion, Argus ML extraction, DARTbot research, Spotlight analytics/benchmarking, iConfirm, Trustworthy AI gate
ey_gam_lite.yaml	EY	52	757	955	44,641	Canvas risk/materiality, Helix analytics, Atlas methodology, specialist coordination, EQR, digital audit, GAM compliance

Every step is annotated with:

• judgment_level: data_only (fully automatable), ai_assistable (AI drafts, human reviews), or human_required (professional skepticism needed)
• ai_capabilities: what AI tools/techniques apply to this step
• human_responsibilities: what the auditor must personally decide or approve

Judgment Level Distribution

Firm	Data-Only	AI-Assistable	Human-Required
KPMG Clara	11%	13%	75%
PwC Aura	11%	13%	76%
Deloitte Omnia	12%	14%	74%
EY GAM Lite	11%	14%	74%

Note: These blueprints are derived from public ISA standards with firm-specific flavours based on publicly available documentation. They are not scraped from proprietary firm methodologies and should be understood as ISA-based approximations.

Generic Audit Blueprints

Blueprint	Framework	Phases	Procedures	Steps	Description
generic_fsa.yaml	ISA	3	9	24	Compact Financial Statement Audit
generic_ia.yaml	IIA-GIAS	9	34	82	Internal Audit (96.2% GIAS coverage)

Generation Overlays

Overlays control simulation parameters without changing the methodology blueprint:

Overlay	Revision Probability	Timing (mu hours)	Anomaly Rate	Cost Multiplier
overlay_default.yaml	15%	24h	~5%	1.0x
overlay_thorough.yaml	30%	40h	~2%	1.5x
overlay_rushed.yaml	5%	8h	~15%	0.6x

Key Features

All FSM blueprints include:

• Phase gates with procedure completion requirements
• Precondition DAGs for execution ordering (no cycles)
• 4-state FSM aggregates per procedure: not_started → in_progress → under_review → completed
• Actor roles with firm-specific naming (partner, manager, senior, staff + EQR/QRP/specialists)
• Standards catalog with ISA paragraph-level traceability
• Evidence catalog with document lifecycle tracking (draft → under_review → finalized)
• Decision branching within procedure steps
• Discriminator filtering for scope control (tiers, categories)
• Cost model with base_hours and required_roles per procedure
• Iteration limits configurable per procedure for revision loop control

Usage with DataSynth

CLI Quick Start

# Run a KPMG-style audit engagement
datasynth-data audit run --blueprint builtin:kpmg --overlay builtin:default --output ./kpmg_output

# Validate a custom blueprint
datasynth-data audit validate --blueprint path/to/custom_blueprint.yaml

# Inspect blueprint structure
datasynth-data audit info --blueprint builtin:pwc

# Generate benchmark dataset
datasynth-data audit benchmark --complexity medium --output ./benchmark/

Configuration Reference

audit:
  enabled: true
  fsm:
    enabled: true
    blueprint: builtin:kpmg    # builtin:fsa, builtin:ia, builtin:kpmg, builtin:pwc,
                                # builtin:deloitte, builtin:ey_gam_lite, or file path
    overlay: builtin:default   # builtin:default, builtin:thorough, builtin:rushed, or file path

Programmatic Usage (Rust)

use datasynth_audit_fsm::loader::{BlueprintWithPreconditions, load_overlay, OverlaySource, BuiltinOverlay};
use datasynth_audit_fsm::engine::AuditFsmEngine;
use datasynth_audit_fsm::context::EngagementContext;

let bwp = BlueprintWithPreconditions::load_builtin_kpmg().unwrap();
let overlay = load_overlay(&OverlaySource::Builtin(BuiltinOverlay::Default)).unwrap();
let mut engine = AuditFsmEngine::new(bwp, overlay, ChaCha8Rng::seed_from_u64(42));
let result = engine.run_engagement(&EngagementContext::test_default()).unwrap();

println!("Events: {}, Artifacts: {}", result.event_log.len(), result.artifacts.total_artifacts());

Generation Profiles

Complete DataSynth YAML configurations targeting specific use cases:

Profile	Industry	Purpose
fraud-detection-basic.yaml	Retail	ML fraud model training (5% fraud rate)
data-quality-testing.yaml	Financial Services	DQ tool validation with systematic issues
ml-training-balanced.yaml	Manufacturing	Balanced anomaly detection with drift
process-mining-full.yaml	Manufacturing	OCEL 2.0 event logs for process mining
comprehensive-demo.yaml	Manufacturing	Full-featured demo (3 companies, all features)
external-audit-engagement.yaml	Manufacturing	ISA-based audit lifecycle with workpapers

Explorer UI

An interactive single-page application for browsing and visualizing blueprints.

cd explorer && python3 -m http.server 8080
# Open http://localhost:8080

Features: procedure flow graphs, state diagrams, config tree browser, causal DAG visualization, export (YAML/JSON/Mermaid), dark/light theme.

Contributing

1. Fork the repository
2. Create a feature branch
3. Add your blueprint(s) to the appropriate directory
4. Ensure YAML validates with datasynth-data audit validate --blueprint your_file.yaml
5. Submit a pull request

Blueprint Authoring Guidelines

• All FSM blueprints must follow schema_version: "1.0" format
• Procedures must have unique IDs within a blueprint
• Precondition references must point to valid procedure IDs (no cycles)
• Standards references should cite public standards only (no proprietary content)
• Every step should have judgment_level set for LLM interaction classification
• Include base_hours and required_roles on procedures for cost model support

License

Apache License 2.0 — see LICENSE for details.

All blueprints in this repository are based on publicly available standards (ISA, IIA-GIAS, COSO, SOX) and publicly documented firm platform capabilities. They contain no proprietary methodology content.