CSV is a family of delimiter-separated text formats rather than one universally interpreted grammar. RFC 4180 documents a common comma-and-CRLF shape, yet spreadsheet exports and regional conventions routinely use semicolons, tabs or other separators. A converter should state which dialect it accepts before it maps rows to objects.
Select the character that separates fields
CSVBridge offers four finite choices: comma, semicolon, horizontal tab and pipe. It does not inspect the data and declare one choice likely. That avoids a file containing many commas in prose being misread when its real separator is a semicolon. The chosen character is structural only outside a quoted field; inside quotes it remains ordinary cell text.
Treat CRLF, LF and CR as record endings
Files move between Windows, Unix tools and older producers, so CSVBridge accepts CRLF, LF and lone CR outside quotes. A final line ending closes the last record but does not invent another empty record. Inside a quoted field the same characters are preserved exactly and JSON escaping makes them visible as content rather than splitting one logical record.
Keep one rectangular table contract
The first record supplies exact JSON keys and every later record must contain the same number of fields. CSVBridge does not pad short rows, discard extra cells, trim headers or merge duplicate names. Those repairs can hide an export defect and shift values beneath the wrong key. A visible error with record and field coordinates keeps the structural decision with the user.