When budgets force tradeoffs in which dialects or accents to record first in speech data collection, how do you decide what to prioritize? Can you share one prioritization rule you used in a real project?