--- license: apache-2.0 task_categories: - tabular-classification - tabular-regression - graph-ml - other tags: - synthetic - financial-data - vynfi - group-audit - consolidation - intercompany - audit-analytics - ifrs - ias-21 - ifrs-10 - ias-28 - cta - nci - method-a - accounting-network - graph-neural-network - enterprise-scale size_categories: - 100M **Note:** the consolidated edge list lives inside the tarball > `enterprise_2000.tar.zst` at the path > `enterprise_2000/consolidated/je_network.parquet`. See the per-entity > walkthrough below for partial-download examples. ## Quick start (full archive) ```python from huggingface_hub import snapshot_download import json, pathlib # Note: tarball is ~3 GB compressed; uncompressed is ~71 GB local = pathlib.Path(snapshot_download( repo_id="VynFi/vynfi-group-audit-enterprise-2000", repo_type="dataset")) # Extract: import subprocess subprocess.run(["tar", "-I", "zstd", "-xf", str(local / "enterprise_2000.tar.zst"), "-C", str(local)]) root = local / "enterprise_2000" cfs = json.loads( (root / "consolidated/consolidated_financial_statements.json") .read_text()) print("Group:", cfs["balance_sheet"]["group_id"]) print("Total assets (USD):", cfs["balance_sheet"]["total_assets"]) print("Total L+E+NCI (USD):", cfs["balance_sheet"]["total_liabilities_plus_equity_plus_nci"]) print("NCI separately presented:", cfs["balance_sheet"]["total_nci"]) ``` ## Quick start (per-entity walkthrough) ```python import json, tarfile from huggingface_hub import hf_hub_download # Download just the tarball, then extract one entity's slice in-memory tar_path = hf_hub_download( repo_id="VynFi/vynfi-group-audit-enterprise-2000", filename="enterprise_2000.tar.zst", repo_type="dataset", ) # (Use 'tar -I zstd -xf' on disk for full-archive extraction) ``` ## Schema highlights **Per-entity** (`entities/{code}/`) carry the v5.x single-entity output shape unchanged. See the [VynFi DataSynth README](https://github.com/mivertowski/SyntheticData) for the ~20 typed-snapshot subdirectories. **Per-entity `graphs/je_network.{csv,parquet}`** (NEW v5.10) — 15 columns: ``` edge_id, document_id, posting_date, from_account, to_account, from_line_id, to_line_id, amount, confidence, predecessor_edge_id, business_process, is_fraud, is_anomaly, ic_pair_id, ic_partner_entity ``` **Consolidated `consolidated/je_network.{csv,parquet}`** (NEW v5.10) — 18 columns: ``` edge_id, document_id, entity_code, posting_date, from_account, to_account, from_line_id, to_line_id, amount, confidence, predecessor_edge_id, business_process, is_fraud, is_anomaly, ic_pair_id, ic_partner_entity, is_eliminated, eliminates_ic_pair_id ``` **Group-level files** under `consolidated/` and `ic_eliminations/` match the v5.0 spec §9 schema. Field-by-field documentation lives in the engine's `crates/datasynth-group/src/aggregate/` modules at the v5.10.0 release tag. ## What this dataset is good for - **Audit ML benchmarks** — large-scale group-audit simulation with known ground truth (every fraud / anomaly / IC pair labelled). - **Graph-ML benchmarks (NEW)** — 5.6 M-edge accounting network spanning 2,000 entities with explicit IC pair linkage and elimination-edge labels. See companion model [`VynFi/je-fraud-gnn`](https://huggingface.co/VynFi/je-fraud-gnn) for a single-entity baseline. - **Consolidation engine validation** — drop-in reference for testing custom IFRS / ASC 810 consolidation logic. - **Education** — concrete example of a 2 000-entity multinational consolidation for accounting / audit pedagogy. - **Performance benchmarking** — the engine's published 69 GiB peak / 8-minute wall-clock profile on `Standard_NC40ads_H100_v5` is reproducible against this exact fixture. ## What this dataset is NOT - Real-world data. Every value is synthetic and deterministically generated. Statistical distributions are approximate models, not samples from any specific company. "ACME" is a fictitious name and has no relationship to any real entity. - A research-grade fraud benchmark. Fraud labels are *injected by construction*, not discovered via investigation. - A regulatory filing. The IFRS treatment is faithful to the published standards but the underlying numbers are fictitious; do not use for any compliance purpose. ## License Apache 2.0. Free for commercial use, modification, distribution, private use; see `LICENSE` for the full terms. ## Citation ```bibtex @misc{ivertowski2026datasynth, author = {Ivertowski, Michael}, title = {{DataSynth}: Reference Knowledge Graphs for Enterprise Audit Analytics through Synthetic Data Generation with Provable Statistical Properties}, year = {2026}, month = {April}, howpublished = {SSRN Working Paper}, url = {https://ssrn.com/abstract=6538639} } ``` ## Related VynFi datasets - [`VynFi/vynfi-journal-entries-1m`](https://huggingface.co/datasets/VynFi/vynfi-journal-entries-1m) — single-entity 1 M JE lines + COA + TB + cost / profit centres + Method-A accounting-network edge list - [`VynFi/je-fraud-gnn`](https://huggingface.co/VynFi/je-fraud-gnn) — trained GraphSAGE fraud + GAE anomaly model (companion to the journal-entries dataset) - [`VynFi/vynfi-audit-p2p`](https://huggingface.co/datasets/VynFi/vynfi-audit-p2p) — P2P document-flow corpus - [`VynFi/vynfi-supply-chain-ocel`](https://huggingface.co/datasets/VynFi/vynfi-supply-chain-ocel) — Native OCEL 2.0 event log - [`VynFi/vynfi-aml-100k`](https://huggingface.co/datasets/VynFi/vynfi-aml-100k) — Banking + AML labels - [`VynFi/vynfi-sar-narratives`](https://huggingface.co/datasets/VynFi/vynfi-sar-narratives) — Banking + AML labels + SAR narratives - [`VynFi/vynfi-ocel-manufacturing`](https://huggingface.co/datasets/VynFi/vynfi-ocel-manufacturing) — Lightweight reconstructed-events prototyping companion ## Related VynFi Spaces - 🔗 [`VynFi/accounting-network-explorer`](https://huggingface.co/spaces/VynFi/accounting-network-explorer) — Interactive ISO 21378 account-class graph - 🛡️ [`VynFi/fraud-gnn-demo`](https://huggingface.co/spaces/VynFi/fraud-gnn-demo) — Gradio fraud-GNN inference demo - 📊 [`VynFi/process-mining-demo`](https://huggingface.co/spaces/VynFi/process-mining-demo) — pm4py process-mining showcase