Inconsistent classification and inclusion of maritime vs land boundaries

Below I've rendered the class=maritime boundaries in red and all others (currently always maritime) in blue.

Can we get some insight on how these classes are assigned? Parts of the Hudson River are classified as maritime and another part are classified as land.

Areas like Riker's Island have an insane level of granularity splitting up a the line! It's like there was a pixel detection classifier run over this.

At the moment, the classification seems to be more or less random, and the classifications are not supported by the OpenStreetMap source data. Here is the OSM Way for that Bronx / Queens border. It's a single way, and no tags on the way or any of the parent relations suggest a maritime border.

Meanwhile the Great Lakes are classified as land borders (this is consistent with OSM and how most maps want to render these, even if they don't render more typical ocean maritime borders). So I guess wrapped up in this is a schema question of whether 2 classes are enough.

I also noticed a bunch of external borders, like this one in NJ are missing from the Division boundaries. https://www.openstreetmap.org/way/309513136#map=11/39.0272/-74.7548&layers=N. Typical maritime boundaries like the this one (a way that is part of the US international border and the EEZ) are also omitted. (These also seem worth including with an expanded set of classes.)

Start by tracing how Division boundaries are sourced and how the maritime/land class is assigned, using the Hudson River, Riker’s Island, Great Lakes, and New Jersey examples. Compare the generated boundaries with the linked OSM ways and inspect the current boundary schema. Done means the classification and inclusion rules are explained and the reported inconsistencies or schema gaps are addressed.