Datasets:
| \begin{table*}[t] | |
| \centering | |
| \small | |
| \caption{Diagnostic workload outside the headline tests.} | |
| \label{tab:diagnostic-inventory-v5} | |
| \resizebox{\textwidth}{!}{% | |
| \begin{tabular}{lll} | |
| \toprule | |
| Diagnostic & Scale & What it exposes \\ | |
| \midrule | |
| Protocol baseline rows & 856 & Random/in-city/LOCO split sensitivity \\ | |
| Failure rows & 371 & Where strict transfer produces city-level failures \\ | |
| Directed domain-shift pairs & 14,762 & How far source and target cities differ \\ | |
| City robustness rows & 138 & Which cities improve or regress by target \\ | |
| \bottomrule | |
| \end{tabular} | |
| }% | |
| \end{table*} | |