I agree with the author that the changelog may be more important for humans. So instead of having machine readable version numbers, maybe we should focus on having machine readable changelogs where items belong to a predefined taxonomy? Aggregating changelogs from multiple packages would make it easier to see the potential impact on your solution.