Contamination Map Methodology

Time series / Dataset

TSB-AD-U (TSB-AD univariate)

Univariate part of the TSB-AD benchmark (Liu & Paparrizos, NeurIPS 2024 D&B): 870 curated, labelled univariate time series from 23 datasets, 12 of which are univariate channels extracted from multivariate datasets (e.g. SMD (U), MSL (U), SMAP (U), SWaT (U)); the others include UCR (anomaly archive), NAB, YAHOO, IOPS, MGAB, WSD, SED, TODS, NEK, Stock and Power. Series were filtered and labels checked or corrected (extended/segmented) by human annotators assisted by detectors. Series-level split: Eval set (350 series, with sampling to balance UCR, YAHOO and WSD) and Tuning set (48 series) for hyperparameter tuning. Within each series, the file name gives a training prefix ('tr_N') that semi-supervised detectors may train on; labels cover the whole series.

Also known as
TSB-AD-U, TSB-AD univariate, TSB-AD-U-Eval, TSB-AD-U-Tuning
Domain
mixed
License
Apache-2.0 for the preprocessing/curation; each source dataset keeps its own licence
Test items from
2024-11-20
Evidence
“[Nov. 20, 2024]”
raw.githubusercontent.com · readme · News · retrieved 2026-10-15
“is online and release TSB-AD-U/M dataset.”
raw.githubusercontent.com · readme · News · retrieved 2026-10-15
Homepage
thedatumorg.github.io

Related datasets

Exposure by model

BenchmarkModelLevelReason
Time-series anomaly detection Chronos-2-Synth Not seen absent from the complete list of training data
Time-series anomaly detection Chronos-2 Not seen absent from the complete list of training data
Time-series anomaly detection Chronos-Bolt Unknown training data not disclosed in enough detail
Time-series anomaly detection Chronos (T5) Not seen absent from the complete list of training data
Time-series anomaly detection DADA (General Time Series Anomaly Detector with Adaptive Bottlenecks and Dual Adversarial Decoders) Test data seen a contamination analysis found test examples in the pre-training data (Third-party (TimeRCD authors): DADA's pre-training corpus (IOPS, MGAB, YAHOO) heavily overlaps their zero-shot test benchmarks drawn from TSB-AD.) [third-party study]
Time-series anomaly detection Granite-TimeSeries-TTM-R1 Not seen absent from the complete list of training data
Time-series anomaly detection Granite-TimeSeries-TTM-R2.1 Not seen absent from the complete list of training data
Time-series anomaly detection Granite-TimeSeries-TTM-R2 Not seen absent from the complete list of training data
Time-series anomaly detection Granite-TTM-R3 Unknown training data not disclosed in enough detail
Time-series anomaly detection Lag-Llama Not seen absent from the complete list of training data
Time-series anomaly detection Mantis (Mantis-8M) Unknown training data not disclosed in enough detail
Time-series anomaly detection Mantis+ (MantisPlus) Not seen absent from the complete list of training data
Time-series anomaly detection MantisV2 Not seen absent from the complete list of training data
Time-series anomaly detection Moirai-1.0-R Not seen absent from the complete list of training data
Time-series anomaly detection Moirai-1.1-R Unknown training data not disclosed in enough detail
Time-series anomaly detection Moirai-2.0-R Unknown training data not disclosed in enough detail
Time-series anomaly detection Moirai-MoE-1.0-R Not seen absent from the complete list of training data
Time-series anomaly detection MOMENT-1 Related data seen related dataset UCR Time Series Anomaly Archive (Hexagon ML/UCR, KDD Cup 2021) was used (via Time Series Pile)
Time-series anomaly detection NuTime Not seen absent from the complete list of training data
Time-series anomaly detection Sundial (Timer 3.0) Unknown training data not disclosed in enough detail
Time-series anomaly detection Time-MoE Unknown training data not disclosed in enough detail
Time-series anomaly detection Time-RCD (TimeRCD) Not seen absent from the complete list of training data
Time-series anomaly detection Timer-S1 Unknown training data not disclosed in enough detail
Time-series anomaly detection Timer-XL (timer-base-84m) Unknown training data not disclosed in enough detail
Time-series anomaly detection Timer Not seen absent from the complete list of training data
Time-series anomaly detection TimeRadar Unknown training data not disclosed in enough detail
Time-series anomaly detection TimesFM 1.0 Unknown training data not disclosed in enough detail
Time-series anomaly detection TimesFM 2.0 Unknown training data not disclosed in enough detail
Time-series anomaly detection TimesFM 2.5 Unknown training data not disclosed in enough detail
Time-series anomaly detection TimesFM 3.0 Unknown training data not disclosed in enough detail
Time-series anomaly detection TiRex 1.1 (gifteval) Unknown training data not disclosed in enough detail
Time-series anomaly detection TiRex-2-fevbench (TiRex-2-f) Not seen absent from the complete list of training data
Time-series anomaly detection TiRex-2-gifteval-pretrain (TiRex-2-gp) Unknown training data not disclosed in enough detail
Time-series anomaly detection TiRex-2-gifteval-zs (TiRex-2-g) Not seen absent from the complete list of training data
Time-series anomaly detection TiRex-2 Unknown training data not disclosed in enough detail
Time-series anomaly detection TiRex Not seen absent from the complete list of training data
Time-series anomaly detection Toto-Open-Base-1.0 Unknown training data not disclosed in enough detail
Time-series anomaly detection Toto-2.0-2.5B-FT Unknown training data not disclosed in enough detail
Time-series anomaly detection Toto 2.0 Unknown training data not disclosed in enough detail
Time-series anomaly detection Toto (2024 technical report) Unknown training data not disclosed in enough detail
Time-series anomaly detection TSPulse r1 Not seen the authors state the dataset was excluded from training
Time-series anomaly detection TTM-R3 Unknown training data not disclosed in enough detail
Time-series anomaly detection Tiny Time Mixer (TTM) Research-Use R2 Not seen absent from the complete list of training data
Time-series anomaly detection UniTS few-shot anomaly detection checkpoints Related data seen in fine-tuning data: related dataset Server Machine Dataset (SMD) was used
Time-series anomaly detection UniTS Not seen absent from the complete list of training data

Notes

published_from = README news item announcing the release of TSB-AD-U/M (the paper was reviewed on OpenReview earlier in 2024; the data download is https://www.thedatum.org/datasets/TSB-AD-U.zip). The public zip redistributes curated series of the source datasets, including SWaT, whose official copy requires a request form. File-list counts: 870 series in TSB-AD-U.csv, 822 in TSB-AD-U-Eva-Full.csv, 349 in TSB-AD-U-Eva.csv, 47 in TSB-AD-U-Tuning.csv (paper: 350 Eval / 48 Tuning). Other TSB-AD-U sources (IOPS, MGAB, WSD, SED, TODS, NEK, Stock, Power, Daphnet, CATSv2, LTDB, TAO, Exathlon, MITDB, SVDB, OPPORTUNITY) have no registry id yet.

Sources