SEP 9, 2026 · PREPRINT
A Later Test Set Is Not a New Domain: Pretraining Familiarity Survives a Contamination-Free Hold-Out
arXiv
A methodologically sound empirical study of pretraining data leakage in time-series models, but limited to a single hold-out dataset, lacks peer review, and reports an important negative result about what does not predict performance without establishing causal mechanisms.
Reported
MASE improvement on weekly Wikipe…28% lower than best classical method
TimesFM vs Chronos rank differenc…−0.53 ranks (1,500 series)
TimesFM vs Chronos rank differenc…−0.09 (754 series)