The Newberry Library dataset we imported was in many ways optimized for visualizations at smaller scales than we increasingly care about as we turn our attention to more local boundaries and other features. They also optimized the dataset to be viewed as an isolated layer rather than integrated into a well-rounded historical dataset like we’re trying to build. As a rule, they didn’t include any large bodies of water in county boundaries, which sometimes resulted in feature metadata noting discrepancies like certain boundary changes being totally “unmappable”.
In Anglo-American legal systems, whether a jurisdictional boundary changes with a waterway depends on the language of the boundary description and the circumstances surrounding the change. Courts seek to distinguish avulsion from accretion, which doesn’t always come easy.
To the extent you’re able to track down sources explaining any notable shifts, you can feel free to record them and reflect them in the boundary. But it’s also acceptable to temporarily skip some details for the sake of keeping the boundary manageable. Make sure to note any errata in note=* and fixme=* tags, akin to what the Newberry Library did. In time, we may have a reason to track certain boundaries more granularly based on accretion, just as we track some boundaries during wartime based on shifting front lines.