Benchmark RadarTS-Fault
Evaluates time series forecasting models under four explicit structural fault modes (time-warped shock, dependency-fracture shock, regime-transition missingness, cascading sensor-to-system failure) injected into lookback windows, with paired clean/corrupt protocol and five difficulty levels across nine datasets and six domains.
- Released
- 2026-06-16
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Standard clean-data leaderboards assume a single error metric predicts deployed reliability, but real faults are structured events. TS-Fault provides a diagnostic protocol that isolates robustness to named fault mechanisms at tunable severities, revealing that clean accuracy anti-correlates with robustness and mechanism-level faults reorder model rankings.
Motivation
Time series forecasting (TSF) underpins consequential decisions in energy, transportation, finance, and healthcare, yet TSF models are almost universally ranked by a single number (e.g., average error) on clean held-out data, under the implicit assumption that it predicts deployed reliability.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.