The metastatistical extreme value approach proved promising in the frequency analysis of daily precipitation from ordinary events, outperforming traditional methods based on sampled extremes. However, sub-daily applications are currently restrained as it is not known if ordinary events can be consistently examined over durations, and it is not clear to what extent their entire distributions represent extremes. We propose here a unified definition of ordinary events across durations, and suggest the simplified metastatistical extreme value formulation for dealing with extremes emerging from the tail, rather than the entire distributions, of ordinary events. This unified framework provides robust estimates of extreme quantiles (≤10% error on the 100-year from a 26-year long record), and allows scaling representations in which ordinary and extreme events share the scaling exponent. Future applications could improve our knowledge of sub-daily extreme precipitation and help investigating the impact of local factors and climatic forcing on their frequency.