well heres how i think about DYN
vocaloid is a collection of random sounds
arrange them together into a melody, but its still random syllables strung together, the syllables dont know what each other are
people speak or sing with a certain volume for each syllable, as they approach the sentence or melody.
with the vocaloid, the syllables each have a certain volume as they are recorded, it will flow better if you bring the DYN (usually downward) of each syllable to match your singing or recording reference.
I can talk for pages and pages of how I approach tuning for each parameter. Oh, and I disclaim knowing whats right since Im mostly self taught. Not sure about everyone else