Select the current interpretation before counting historical human prompts
For historical coding-agent comparisons, first build a source union, then choose one interpretation per source message, and only then filter for human-authored activity. Counting a legacy message table alone can omit continuing work; summing versioned projections can simultaneous…