How the score works
For each candidate (comma, semicolon, tab, pipe, colon, caret) the sample is parsed completely, respecting quotes, so a comma inside a quoted field does not count. The score combines the share of records that have the most common width, a preference for more than one column, the rate of quote anomalies and the sample size.
Confidence compares the best score with the runner-up. In the example above every line contains one comma and one semicolon, so both produce two consistent columns. The detector reports roughly 50% confidence and asks you to choose instead of pretending to know.
Things that are not detection failures
- A one-column file: no delimiter splits it, and that is a valid result.
- A header-only file: one record and no data is reported as such.
- An empty file: zero records, no crash, no error.
Header, quote and line-ending evidence
The same pass suggests a header by comparing the first record with the column types below it, reports duplicate or empty header names, flags a possible multi-row header, and counts CRLF, LF and CR terminators so mixed line endings are visible before you parse.
Questions
Why not always pick the most frequent character?
Frequency is misleading: prose fields contain commas, decimal numbers contain dots and commas, and quoted fields can contain anything. Consistency of record width under a real parser is much stronger evidence.
Can I use a delimiter that is not listed?
Yes. Type any single character in “Other”. Multi-character delimiters such as “||” are a separate profile that is not supported yet; they are refused rather than silently reduced to one character.
Does changing the delimiter modify my file?
No. Every change reparses the original bytes from the start.