One grid, nine table formats

TSV if the data is going straight into a spreadsheet by paste, CSV if it is going in as a file. A tab-separated block pastes into Excel, Google Sheets and Numbers without an import dialog, because tabs are what those applications put on the clipboard themselves. For files, note that Excel on a machine set to a European locale reads a comma-separated file as one column and expects semicolons instead — the separator control covers that case, and it is the single most common reason a CSV opens looking broken.

They are not, unless you untick “Numbers and booleans unquoted”. With it on, a cell matching an integer, decimal or exponent form becomes a JSON number, true and false become booleans, an empty cell becomes null, and everything else stays a string. Turn it off when the values only look numeric — a zip code, a phone number, an order reference with leading zeros — because JSON numbers cannot preserve the leading zero and will silently drop it.

The quoting is correct: an apostrophe inside a value is doubled, which is the SQL standard escape, an empty cell becomes NULL rather than an empty string, and column names are lowercased and stripped to letters, digits and underscores so they are valid identifiers. What the generator cannot know is your schema — every column in the optional CREATE TABLE is typed TEXT, there are no keys or constraints, and nothing is parameterised, so treat the output as a seed script or a fixture rather than something to point at production.

Only for the default style: \toprule, \midrule and \bottomrule come from booktabs, so add \usepackage{booktabs} to your preamble. Untick it and you get plain \hline rules with vertical bars in the column specification, which needs nothing at all. Cell text is escaped for the ten characters LaTeX treats as special — ampersand, percent, dollar, hash, underscore, braces, tilde, caret and backslash — so a value like 50% or user_id compiles instead of erroring.

Because the columns are padded by counting characters, and some characters are drawn two cells wide. CJK ideographs, most emoji and a few symbols occupy two terminal columns while counting as one character, so a column containing them ends up one cell short per occurrence. There is no fix that works everywhere — terminal width tables disagree between emulators — so if your data is Latin text the borders line up exactly, and if it is not, the plain + - | style tends to look less broken than the box-drawing one.

That page does one format properly and this one does nine adequately. If the destination is a README or a pull-request comment, use the Markdown table generator: it shows the table the way a repository host would render it, offers padded and compact pipes, and re-imports an existing table for editing. Come here when the same rows have to leave as something else — an HTML table for a page, INSERT statements for a fixture, LaTeX for a paper, or a box-drawn table for a terminal help screen.

Press Flip axes and the grid is transposed — the header row becomes the first column and each old column becomes a row. It is the fastest fix for data that arrived the wrong way round, which happens constantly with exported metrics where every series is a row and every date is a column. The alignment settings reset when you flip, because they belonged to the old columns.