I’ve been getting better fits on small (8–15 km2) urban watersheds by calibrating the baseflow recession k first, then touching CN only after NSE clears about 0.8 on low-flow segments. With 5-min rain and 15-min flow, HEC-HMS Optimization Trial runs clean if I fix initial baseflow and bound k to 0.90–0.98, cutting the loop to about 30 minutes — anyone else front-load k like this or see quickflow bias?
But i front-load k too on small urban basins; I pull a couple dry-weather recessions, fit ln(Q) vs time to get k, then seed that so HMS converges faster — like giving the optimizer a head start. Small caveat for @OP: in leaf-off or cold snaps, that “k to 0.90–0.98” box has missed a few events for me (needed about 0.86), so I widen bounds or use Log NSE to weight lows. HMS docs on objective functions are handy if you go that route: HEC-HMS Documentation.
I’ve had good luck telling HMS to optimize on NSE (log) for the low-flow pass and giving it a 2–3 day warm-up, then flipping back to NSE once k settles; with 15-min flow I keep k in the “0.90–0.98” window like you and it trims the run to about 30 minutes. Small caveat: watch CN creeping to cover baseflow error if the pervious slice is tiny.