Quantitative early-warning performance of CT-eProber on the HUST-LIB dataset. (a) Cycle-by-cycle prediction of remaining useful cycles (RUC) to the knee point: predicted versus actual values for 22 test cells. Results for intermediate states: (b) State-of-health (SOH) estimation at the current cycle (${\text{SOH}}_{\text{C}}$) versus cycle number, and (c) SOH prediction over next 10 cycles (${\text{SOH}}_{\text{F}}$) versus cycle number. The predicted ${\text{SOH}}_{\text{F}}$ values are further fed into the post-event quantile regression to detect the knee point in real time, providing a secondary verification step that enhances early-warning reliability. For clarity, results in (a)–(c) are shown at 10-cycle intervals. (d) Relative-error density distributions for ${\text{SOH}}_{\text{C}}$ and ${\text{SOH}}_{\text{F}}$. (e) MAPE of predicted RUC for each test cell.