Make a Safe Data Sample to Attach in Public

Repro sample for a CSV parser bug

The bug is an unescaped quote in the amount column on row 2. Names/emails get faked, but the amount value (the actual cause) and the id are preserved, so the sample still breaks the parser.

Input (repro.csv):
id,name,email,amount
1,Sarah Chen,sarah.chen@acme.io,"1,200"
2,Tomás Reyes,treyes@globex.com,"3""500"

Output (repro-scrambled.csv):
id,name,email,amount
1,Dr. Elena Rosales,Reanna.Lockman@yahoo.com,"1,200"
2,Marcus Hettinger,Jaylin.Bode@gmail.com,"3""500"

Names/emails faked; the malformed amount that triggers the bug
and the row ids are preserved exactly.

JSON sample for a GitHub issue

A nested record reproducing a serialization bug. name/email are faked; the id, the numeric edge value 0, and the empty-array tags (the actual repro) survive.

Input (issue.json):
{
  "id": "ord_8821",
  "customer": { "name": "Priya Nair", "email": "priya@shop.co" },
  "qty": 0,
  "tags": []
}

Output (issue-scrambled.json):
{
  "id": "ord_8821",
  "customer": { "name": "Dr. Elena Rosales", "email": "Reanna.Lockman@yahoo.com" },
  "qty": 0,
  "tags": []
}

Open dataset row sample

Publishing rows of an analytics dataset. Identifier columns are faked; the measured columns the dataset is actually about are preserved, so the published sample is both useful and safe.

Input (sample.csv):
name,email,page,duration_ms,bounced
Li Wei,li.wei@x.com,/pricing,4200,false

Output (sample-scrambled.csv):
name,email,page,duration_ms,bounced
Mavis Goldner,Lonnie_Cremin@hotmail.com,/pricing,4200,false

The analytic columns (page, duration_ms, bounced) survive;
only the identifiers were faked.

Reproducible sample for a long thread

Set a seed so you can hand the maintainer the exact same anonymised file again later without re-leaking anything.

Input (case.csv):
case_id,first_name,last_name,email,status
9,Aisha,Khan,aisha.khan@corp.net,open

seed = 99

Output (every run with seed 99 is identical):
case_id,first_name,last_name,email,status
9,<fake>,<fake>,<fake>,open

case_id + status preserved -> the maintainer can match the row.

PII inside a message body still needs scrubbing

The tool fakes whole cells in matched columns; it does not scan free text. An email or phone inside a message column would be published as-is because message isn't a PII token. Scrub that column before posting.

Input (log.csv):
id,email,message
5,dana@x.com,"User wrote: call me at 415-555-0199"

Output (log-scrambled.csv):
id,email,message
5,Hilbert.Klein@gmail.com,"User wrote: call me at 415-555-0199"

The email column is faked; the phone inside message is NOT.
Run message through email-phone-scrubber first so it becomes
[REDACTED_PHONE] before you attach the file.

The tool fakes whole cells in name-matched columns and never scans cell contents. An email, phone, or SSN written inside a free-text column (message, log, notes, description) would be published verbatim because the column name isn't a PII token. Scrub those columns first with email-phone-scrubber, which emits fixed [REDACTED_*] tags, before you attach anything in public.

Headers like email_address, customer_name, or home_phone are not exact PII tokens, so the anchored regex leaves them — and real PII — in the file you post. Rename them to the bare token (email, name, phone) before scrambling, and eyeball the output before publishing.

If the value that triggers your bug sits in a column the regex matches (say a malformed name), scrambling will replace it and your sample may stop reproducing. Rename that column to a non-PII header before scrambling so the bug-triggering value is preserved, or reproduce the bug with the value moved to a clearly non-PII column.

In-browser the JSON path runs only when the filename ends in .json. A JSON snippet saved as sample.txt is parsed by PapaParse as a one-column CSV and barely changes. Rename to .json, rely on the server-safe auto sniff of a leading [/{, or set format: json.

If the very bug you're reporting is invalid JSON, JSON.parse will throw and produce nothing — you can't scramble a document the parser rejects. In that case anonymise a valid version, or share the malformed bytes as a CSV/text attachment after manually replacing the PII. CSV is more forgiving and PapaParse parses ragged rows.

ssn / tax_id columns become faker.string.numeric(9) — nine random digits, no dashes, no checksum. Safe to publish, but if your repro depends on a specifically formatted SSN, note that the fake won't match NNN-NN-NNNN; the tool has no SSN-format option.

A header-only CSV returns just the header; a sample with no PII-named columns returns with itemsRedacted = 0 and every cell preserved. No error — the zero count just tells you nothing matched, which is fine for a sample that legitimately has no PII.

A seed lets you regenerate the identical anonymised sample later, but it is not a key and there's no mapping back to the real values. The operation is one-way — keep your original sample locally if you ever need the real data for your own debugging.

Repro samples are usually tiny, but if you attach a big export it must fit the security-family limits: Pro 100 MB / 5 files, Pro-media 500 MB / 50, Developer 2 GB / unlimited (this tool needs at least Pro). Trim to a minimal repro — smaller samples are better for the maintainer and for privacy anyway.