When a datapack with custom biomes is loaded, the biomes it defines are assigned IDs according to their lexicographic order. When a new biome is added to an existing world but its name gets sorted anywhere else than at the end, it causes the biomes that follow it to shift. Chunks only store the resulting IDs in their biome maps and not the names, so they become corrupted, showing the biome that precedes the correct one.

This can be quite easily fixed if chunks stored not only the biome IDs, but also a mapping that allowed retrieving the actual name, kinda like the block palette already works for a chunk section. Since there is usually only one biome per chunk, it's just a matter of storing one additional pair in the chunk data. This would also make it possible to define more than 256 biomes for a world.