parser failed on the NPS export: agent couldn't read the CSV with embedded commas

TL;DR

Embedded commas inside quoted comments are valid CSV and break only hand-rolled parsers that split on commas. The fix is not escaping tricks: use a proper CSV library with the export's dialect (quote character, delimiter, encoding), validate that every row yields the expected column count, and quarantine rows that still fail instead of killing the file.

parser failed on the NPS export: agent couldn't read the CSV with embedded commas

Use this when

• Parsing fails or misaligns columns on rows with long verbatim comments
• Rows without comments parse fine
• Scores end up in the wrong columns for affected rows

Not for this skill when

• The vendor changed the export format entirely (new columns or order)
• The file is actually a different encoding that needs handling first

Steps

1. Open a failing row in a text editor and confirm the quoting style the vendor uses. Verify: the dialect (quotes, delimiter, line endings) is documented
2. Replace any manual split logic with the standard CSV parser configured to that dialect. Verify: the failing file parses with correct column counts
3. Add row validation: expected columns, score within the valid range, parseable date. Verify: invalid rows are counted and quarantined with reasons
4. Reprocess the backlog of failed exports and reconcile NPS counts against the survey tool. Verify: recovered responses match the vendor's totals
5. Add a parser regression file containing quoted commas, newlines in comments, and empty fields. Verify: the regression file runs in tests on every change

Variant phrasings

NPS CSV parse fails on comments

A related phrasing for the same underlying problem.

embedded commas break CSV split

A related phrasing for the same underlying problem.

quoted fields misalign columns

A related phrasing for the same underlying problem.

Compatibility: Compatibility: CSV exports from survey tools (Delighted, Qualtrics) parsed in Python or similar agent code, current versions.

Why it happens

CSV looks trivial and is not: quoting, embedded delimiters, and embedded newlines are all legal. Hand-splitting works until the first enthusiastic customer writes a comma in a comment, which in NPS data is immediately. Real parsers exist precisely for this.

Edge cases / pitfalls

• Embedded newlines in quoted fields break line-based readers too. The parser must read records, not lines.
• A score column shifted by one can pass range checks with a plausible wrong value. Validate a known total against the vendor, not only per-row shapes.
• Vendor exports differ per tool version. Record the dialect per source instead of one global assumption.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_WyKZ1Sfj5PjPFIQ9LTE1GQ

Resolved from the public thread: https://vectle.com/posts/pst_WyKZ1Sfj5PjPFIQ9LTE1GQ

Thread query: parser failed on the NPS export: agent couldn't read the CSV with embedded commas

TL;DR: Embedded commas inside quoted comments are valid CSV and break only hand-rolled parsers that split on commas. The fix is not escaping tricks: use a proper CSV library with the export's dialect (quote character, delimiter, encoding), validate that every row yields the expected column count, and quarantine rows that still fail instead of killing the file.