natnitaract/AMLworldCompactEval

AMLworld-Compact

Importance-weighted transaction graphs for LLM evaluation and error diagnosis.

Paper · Code & evaluation · Baseline results

Release 1.0 (2026-09-23).

AMLworld-Compact provides 6,021 evaluation cases for anti-money-laundering research, derived from the HI-Small and LI-Small temporal test splits of IBM's synthetic AMLworld dataset. Each case contains a target transaction's local graph as text, an illicit/benign label, an optional laundering typology, and an importance weight for estimating full-split metrics.

Configuration	Evaluation cases	Illicit cases retained	Full test edges	Reduction
HI-Small	3,753	1,251	1,015,669	271×
LI-Small	2,268	756	1,384,810	611×

Both configurations contain a single test split. All illicit edges are retained; benign edges are sampled using supervised difficulty strata. The released set has one illicit case for every two benign cases.

Each row carries case_id, center_edge_id, subset_index, label (with the equivalent illicit flag), typology and typology_id (0–7, or −1 when unannotated), ht_weight, n_transactions, n_tokens_approx, and typed_graph_text. Typology IDs 0–7 correspond to fan-out, fan-in, cycle, scatter-gather, gather-scatter, stack, bipartite, and random, in that order.

Quick start

pip install datasets scikit-learn

from datasets import load_dataset

ds = load_dataset("natnitaract/AMLworldCompactEval", "HI-Small", split="test")
case = ds[0]
print(case["case_id"])
print(case["typed_graph_text"])

Use "LI-Small" to load the other configuration.

Use typed_graph_text as the graph input and keep label, illicit, typology, and typology_id for scoring. Task instructions and few-shot examples are in the code repository's prompt templates. Each graph covers two hops, with at most 50 neighbours per hop. The dataset viewer may shorten long cells; loading the dataset returns the complete text.

Evaluate predictions

Report Horvitz–Thompson (HT)-weighted precision, recall, and F1 as the primary detection metrics. Given one binary predictions array in dataset row order (1 = illicit, 0 = benign):

from sklearn.metrics import precision_recall_fscore_support

precision, recall, f1, _ = precision_recall_fscore_support(
    ds["label"],
    predictions,
    average="binary",
    sample_weight=ds["ht_weight"],
    zero_division=0,
)
print(f"P={precision:.4%}  R={recall:.4%}  F1={f1:.4%}")

Join external predictions by (configuration, case_id) before scoring: case IDs repeat across configurations. Choose any decision threshold using separate validation data.

Weights account for the sampled benign edges. HT-weighted counts estimate the full temporal test split; precision and F1 are ratios of those counts and are not guaranteed to be unbiased. Unweighted scores describe the compact set's 1:2 class ratio: predicting every case illicit gives 50% unweighted F1. Label these two metric frames separately.

For typology evaluation, use illicit cases with a known ground-truth typology: 791 HI-Small cases and 174 LI-Small cases. A missing typology does not mean benign. State whether the score covers all annotated illicit cases or only correctly detected ones.

The evaluation guide provides commands to score the released ensemble, run LLMs, and evaluate saved predictions.

Files

load_dataset() loads only the evaluation tables, about 31.7 MiB across both configurations.

Directory	Contents
data/<configuration>/	Test Parquet files with graph text, labels, and weights
extras/<configuration>/	Aligned scoring arrays, graph features, and case indices, in evaluation-table row order
ml_baselines/	Supervised checkpoints under weights/ and full-test predictions under test_probs/, five seeds each

The code repository provides loaders and scoring examples for the arrays and checkpoints.

Models

The primary ML reference averages LightGBM+GFP and XGBoost+GFP, both trained with temporal supervision. The construction ensemble also included GCPAL+GFP, whose random fine-tuning split overlaps roughly 60% of the temporal test edges; it defines the frozen sampling design and is excluded from the primary evaluation. The seven evaluated LLMs are listed in the code repository's baselines table.

Uses

This synthetic benchmark supports research on transaction classification, graph-to-text prompting, and error analysis. Use the released test split for evaluation and separate data for training and tuning. Results describe the released AMLworld splits and do not establish performance on real banking transactions.

Licences and source

Derived from IBM AMLworld. The original transaction CSVs are not included.

• Data and features (data/, extras/): CDLA-Sharing-1.0.
• Supervised model parameters and outputs (ml_baselines/): MIT.

See NOTICE.md for attribution and licence scope.

Contact

Open a GitHub issue for questions about the dataset or the code.

Citation

BibTeX

@misc{nitarach2026amlcompact,
  title  = {AMLworld-Compact: Importance-Weighted Downsampling for LLM Evaluation and Error Diagnosis},
  author = {Nitarach, Natapong and Ngampornsukswadi, Phume and
            Taveekitworachai, Pittawat and Nonesung, Surapon and
            Sirichotedumrong, Warit and Halverson, Duncan and
            Pipatanakul, Kunat},
  year   = {2026},
  url    = {https://openreview.net/forum?id=VouFZFf8Ph}
}

Please also cite AMLworld. Its citation is included in the companion README citation section.