How does 'sort' interact with record batches?

[daphnenhuch-at]

I have a table with 10,000 records in it ranging from primary key "0000" to "9999". I'm doing the following query

ctx
        .table("table")
        .await?
        .sort(vec![ident("userPrimarkyKey").sort(true, true)])?
        .window(vec![row_number().alias(DATA_FUSION_ROW_NUMBER)])

My goal is that I will have a fully sorted file sorted by primary key where each fileRowNumber is the index of that row in the file. However, I'm getting the result where the record batches are sorted, but not with respect to each other. The table returned here is
user primary key 8192 through 9999 followed then by 0 through 8191. The fileRowNumbers are correct in that they actually match what would have been the fully sorted user primary key order. So why are the record batches not returned sorted with respect to each other?