Research Guides
Before Opening a Research CSV, Decide Which Fields Are Text
A research export may contain page titles, customer comments, URLs and measurements. Those fields do not all have the same meaning. Before handing a CSV to a colleague, define how each column should be interpreted rather than assuming the spreadsheet will preserve your intent.
Why a readable export needs a safety check
OWASP describes CSV formula injection: untrusted exported content can be interpreted as a spreadsheet formula. It also warns that quoting and escaping are not universally reliable, particularly across save-and-reopen cycles. A protection suitable for a human-viewed spreadsheet can change data consumed by another program. There is no universal sanitization strategy for every spreadsheet and downstream consumer. OWASP’s CSV injection guidance
This is a reason to test the intended workflow, not a claim that a particular Colquhoun export is vulnerable or already protected.
Write a small export contract
For a competitor-review dataset, agree on a record like this:
- Source URL: text; preserve the original destination.
- Review excerpt: untrusted text; never an intended calculation.
- Rating: a number with a documented scale; missing is not zero.
- Capture date: an explicitly documented date format and time zone where applicable.
- Analyst note: text; separate it from the quoted review.
Keep an unchanged source copy for provenance. Label any viewing-oriented transformed copy so someone does not mistake it for the original input to a later analysis.
Test the handoff, not just the download
Use a disposable test file containing fictional, harmless text. Include punctuation, quotation marks, line breaks, non-English characters and a harmless formula-like string such as =1+1. Test with the exact spreadsheet application and import settings the recipient intends to use. Do not open an untrusted real export merely to see what happens.
Record whether text remains text, whether fields shift into new columns, and whether the result changes after saving and reopening. For programmatic reuse, separately verify that the parser preserves the agreed values. A successful test in one application is not proof for another.
A reusable acceptance record
- Export owner and intended recipient.
- File version, capture period and source dataset.
- Column names, types and missing-value rules.
- Intended application and import procedure.
- Test cases and observed results.
- Any transformations and their effect on original values.
- Unverified applications or workflows.
If the workflow fails, keep the export out of the sharing queue until the responsible developer resolves it. The completion note should say exactly what was tested, not simply “CSV safe.”