Interesting point. The "active customer" example is the one that sticks, since it's not a data quality failure by any dashboard's definition, it's three teams agreeing on a column name while disagreeing on what it measures. In research I've done on how incidents get reported internally, nobody opens a ticket saying "data state changed," they say "the model got worse" or "the number looks wrong," and the diff only gets named after someone traces it back manually. Have you seen a case where a team caught this before the output shifted, rather than after someone already flagged the result as broken?