A single headline Sharpe of 1.2 can hide a strategy that ran at 2.5 for three years and 0.1 for the following five. The rolling chart makes that visible immediately, and it is the first thing to plot when someone shows you a track record.
Choose the window with the noise in mind. A 12-month rolling Sharpe on daily data has a standard error around 1.0, so it will swing between -0.5 and 2.5 for a strategy with a true Sharpe of 1.0 purely by chance. Do not read every dip as decay.
Its real use is as a strategy-monitoring tool with pre-set thresholds. Decide in advance what rolling level, sustained for how long, triggers a review, and write it down before you need it.
Related: sharpe-ratio, strategy-monitoring, standard-error, alpha-decay