>
> Actually, one can infer very safely that in this case the shift from
> /ae/ to /a/ took place in Latin:
that is correct. the only explanation which will satisfy the reality of
"a" is that the change took place in Latin.
the dissimilation did not happen in Rum as expected. For this see beside
the example with "era" the next one:
Latin "aetatis" and Rom. "etate".