How can I detect short or extra fields in CSV rows?

Use Python’s csv.reader to parse the file, count the header fields, and compare that count with every later row. The reader returns each parsed record as a list of fields, so len(row) gives you the count to check. [S1]

import csv

path = "input.csv"  # Replace with your file path.

with open(path, newline="", encoding="utf-8") as file:
    reader = csv.reader(file)
    header = next(reader, None)

    if header is None:
        print("The file is empty.")
    else:
        expected = len(header)
        print(f"Expected fields per data row: {expected}")

        for record_number, row in enumerate(reader, start=2):
            if len(row) != expected:
                print(
                    f"Record {record_number}: expected {expected} fields, "
                    f"found {len(row)}; source line(s) through {reader.line_num}"
                )

This assumes the first parsed record is a header. The script reads and reports; it does not write to the file. The encoding value is an example—use the encoding appropriate for your file.

A comma in a CSV file is not always a field boundary. For example, a properly quoted field might contain a comma: "Miller, Ana" can be one field. Quoted fields can also contain line breaks. The CSV parser applies delimiter and quote settings, whereas splitting raw text on commas cannot reliably distinguish those cases. [S1]

There are two useful numbering ideas in the example. record_number counts parsed records, with the header as record 1, so the first data record is 2. reader.line_num is the number of source lines consumed so far; it is not the parsed record number. A quoted record spanning multiple physical lines can make those counts differ. [S1]

The message says “source line(s) through” to avoid implying that every record occupies exactly one physical line. If you need to identify a record by position in the parsed CSV, use its record number; if you need a physical-line reference, the reader’s line counter provides the number of lines consumed.

Suppose, as an example, the header has three fields and the first data row has two. The report could look like this:

Expected fields per data row: 3
Record 4: expected 3 fields, found 2; source line(s) through 4
Record 7: expected 3 fields, found 4; source line(s) through 7

Those numbers are illustrative, not results from a real file. In the code, next(reader, None) consumes the first parsed record as the header. That is appropriate when your CSV has a header row. If your file has no header, do not treat its first record as column names: set the expected field count from a known schema or another explicit assumption, then check every parsed row starting with the first.

If the file may use a different delimiter or quote style, configure the reader’s dialect to match the file. The example does not infer a schema or determine whether a row’s values are meaningful; it only detects a difference in field count. [S1]

Keep validation separate from any cleanup step. Review the reported record and line references against the original file, then decide what correction is appropriate. This example opens the input for reading and prints findings; it contains no write operation, so it does not repair, remove, or rewrite rows.

Before relying on a report, confirm that the correct file was selected, that the header assumption is right, and that the parser settings fit the CSV. A field-count match is not proof that every value is valid: a row can have the expected number of fields and still contain missing, misplaced, or unexpected content. The check establishes only whether each parsed row’s field count matches the header’s count.