- If the area is shown on two pages, in the case of TPZ you get two features instead picking the best of the two
- In the case of A4Areas and CA you get a multipolygon with both copies unioned together
We need to update the initial schema extraction prompt to produce a single page if there are multiple copies.
Maybe update the 'map_pages' to be a structured json rather than a list of ints?
{"type": "single_page", "best_map_page_number": 2, "other_map_page_numbers": [3] }
or
{"type": "split_across_multiple", "page_numbers": [2,3]}
We need to update the initial schema extraction prompt to produce a single page if there are multiple copies.
Maybe update the 'map_pages' to be a structured json rather than a list of ints?