The context-specific proportion switch (CSPS) effect refers to the modulation of switch costs by the frequency of switch trials within a given context, with larger switch costs in mostly repetition than in mostly switch contexts. Research on the analogous context-specific proportion congruency (CSPC) effect has shown that the extent of this context-specific adaptation depends on the volatility of context changes: the CSPC effect was stronger in a blocked manipulation with a single context change than in a random trial-by-trial manipulation, where stimuli appeared unpredictably across contexts (Fischer et al., 2014). In two studies, we examined whether a similar difference between blocked and random conditions would also emerge for the CSPS effect. Results revealed a typical CSPS effect in the blocked condition but a descriptively reversed effect in the random condition. Unlike conflict, which depends on the congruency of the current trial, task switching relies on a sequential reference to the previous trial to determine switches or repetitions. This dependency becomes critical when the volatility of the context changes increases: the reference stems from the previous context, while the current trial requires adaptation to the other context. Post hoc analyses excluding context-switch trials suggest that adaptation is not fully established on the first trial after a change, potentially explaining the reversed CSPS pattern in the random condition. Overall, these findings suggest that the CSPS effect is selectively influenced by both the volatility of context changes and sequential reference, which must be resolved before effective adaptation to a context change can occur.