Tabulay wants CSV cleanup to leave no surprises

Field notes

The page explains its indexing approach as recording where rows begin, then parsing rows when they are shown or queried. It also describes checking inferred column types across every row, with mismatches counted in the header and examples shown in issue cards. Performance figures are presented as measurements on a 1.3-million-row file; the page says files above 2 GB can still be opened, but only read-only.