How source cells are read

CSV cells begin as text. MLabel reads each value using the input field’s declared type.

Accepted values

CSV cells are strings. Each is coerced according to its declared type when the file loads.

Type	Accepts
text	Anything; non-strings are stringified.
number	Anything Number() parses, after trimming.
integer	The same, but rejects a fractional result.
boolean	true false 1 0 yes no y n, case-insensitive.
date	ISO-8601 and a range of everyday formats. See Dates.
enum	Must be exactly one of the choice names.
composite	A JSON string at the top level, then recursively coerced.
array of scalars	JSON first; failing that, a stringy list. See Lists.

An empty cell becomes null, for every type. Empty is not zero, not false, and not an empty list — it is the absence of a value, which is what makes required-ness meaningful.

A cell that will not coerce does not refuse the file. The value shows as empty with a warning icon you can hover or focus for the reason, and the labeler carries on — they cannot fix the source data, so it is information rather than a task. See Troubleshooting.

Dates

MLabel reads more than new Date() does, to support common source formats:

Written as	Read as
2026-05-01, 2026/05/01	that calendar day
20260501	that calendar day
01/05/2026	5 January — month first
25/12/2026	25 December — day first, since 25 cannot be a month
2026-05-01 14:30, …T14:30:00	that date and time
…T14:30:00Z, …+02:00	that instant, in the zone given
1767225600 / 1767225600000	Unix seconds / milliseconds

An all-numeric date with both parts under 13 is month-first, matching what JavaScript has always done with 01/05/2026. Use ISO dates such as 2026-05-01 to avoid that ambiguity.

Everything is read and displayed as UTC, including values that name no zone. A value that names no zone has not told MLabel one, and adopting the labeler’s would mean two people in two countries exporting different bytes for the same input file.

A date with no time of day displays as a date; one carrying a time displays with it. MLabel knows which by looking at the value, so a source that says exactly 2026-05-01T00:00:00Z is indistinguishable from one that says 2026-05-01, and displays as a plain date.

Lists that are not JSON

An array of scalars accepts JSON first. If that fails, MLabel falls back to reading the cell as a stringy list: it strips one matched layer of [], () or {}, splits on commas, semicolons, pipes or newlines, and unquotes each item.

a, b, c          →  ["a", "b", "c"]
[1; 2]           →  [1, 2]
'red', green     →  ["red", "green"]

Valid JSON always wins, so ["a, b"] stays a single item. Use JSON when you control the source format; it preserves embedded separators unambiguously.

It applies only to lists of scalars. For an array of object, a map or an object, splitting on commas would quietly mis-assign data rather than fail, so those stay strict JSON and a malformed cell is reported.

One consequence worth knowing: a column read this way and carried into the output with fill: copy is written back as JSON. *-output.* has never promised byte-for-byte fidelity, but a, b in becomes ["a","b"] out.