It is shown that the solution trichotomy described by Watson (1984) applies to a much larger class of minimax problems and has additional robustness properties beyond those discussed by him. We also address a number of practical issues, demonstrating, in particular, that the fixed-interval smoothing algorithm can be used, when finite and possibly nonconsecutive data are available, to obtain the minimax solution, the corresponding component estimate, and the robust mean square estimation error.