FraudTwin documentation#

FraudTwin is a deterministic payment-world simulator for fraud detection, graph analysis, data engineering, and machine-learning experiments.

The same configuration and seed produce the same entities, events, labels, and output fingerprints, so difficult fraud cases become explainable, testable, and reproducible.

Generate your first run

Install and run#

$ poetry install
$ poetry run fraudtwin config validate configs/minimal.yaml
$ poetry run fraudtwin generate configs/minimal.yaml --output-dir ./runs

Start here#

Quickstart

Install FraudTwin, create a small run, and learn what the generated manifest and Parquet files mean.

Quickstart
Tutorials

Learn FraudTwin through short, executable paths grouped by workflow.

Tutorials
How-to guides

Configure runs, build datasets, operate integrations, and solve focused implementation tasks.

How-to guides
Explanation

Understand determinism, lifecycles, point-in-time data, fraud graphs, and data quality.

Core concepts
Reference

Look up the Python API, configuration, CLI, and data contracts.

Python API reference

Need a guided route?#

Use Choose your path if you want a recommended starting point based on your goal or experience. It is an onboarding aid, not a second documentation hierarchy.

How a run becomes an experiment#

configuration → entities → behavior → payments → fraud/workflow → datasets / graphs / benchmarks

Each layer keeps stable identities and causal timestamps. Operational views contain only information available at the selected time; oracle artifacts keep the complete explanation for evaluation and audit.