Thanks, but the explanation still doesn't quite make sense because the discrepancy occurs for historical months in the downloaded data. So even if the CSV only downloads (for example) the top 1000 landing pages, it doesn't explain why the same download showed different data later. The top 1000 landing pages in that period should not have changed.
Anyway I think we will start using the API to extract the data in future as this seems to be more reliable regardless, so thanks for the help.