Testing is unfruitful because even if the unit passes the test, capacitance degradation goes on and unit functionality changes on the field. So you need to come up with actual worst-case capacitance and decide if this is acceptable. Losing some 20-30% of capacitance due to aging is completely normal. If you use a physically large part with low DC bias, that effect might be just another 20-30%, then add 20% for tolerance, and your 100nF might be anything from 40nF to say 150nF. If it's a large-capacitance-in-small-package ran at more than single-digit % of voltage rating, so that DC bias effect is significant, make that from 15nF to 150nF.
Of course sometimes a simple RC filter can be acceptable even if the time constant drops to one quarter, in such cases I use X7R MLCCs as well. An example would be filtering 1MHz PWM to give some DC setpoint with reaction time in seconds.