Utilities
Clean & Anonymize
The twenty things wrong with every export — trailing whitespace, twelve spellings of "California", numbers stored as text, four different date formats — found and fixed in one previewable pass. In the same pass, the personal data in the file is found and masked, hashed or dropped.
Everything this tile does is a rule computed on your own device. There is no model, no cloud service and no network call anywhere in it.
What it accepts. CSV, TSV and other delimited text, Parquet, JSON and NDJSON. Excel workbooks are not read directly by this tile in this release — convert the workbook to CSV with Data Converter first, and the panel will tell you so if you drop one.
How to use it
- Open Clean & Anonymize and add a file.
- DataTray scans the file and lists what it found — both quality problems and personal data.
- Build a stack of operations. The list runs top to bottom, and you can reorder it.
- Preview, then run.
The cleaning operations
| Operation | What it does |
|---|---|
| Trim | Removes leading and trailing whitespace. |
| Collapse whitespace | Turns runs of spaces and tabs into one space. |
| Change case | UPPER, lower, Title or Sentence case. |
| Convert type | Coerce a column to text, whole number, decimal, boolean, date, timestamp or time. Coercing a postcode column to text is the fix for 01234 having become 1234. |
| Normalize dates | Turn several written date formats into one ISO date. |
| Normalize numbers | Strip currency symbols and thousands separators, and read US or European decimal conventions. |
| Find and replace | Literal or regular expression, in one column. |
| Split a column | By a delimiter, by fixed widths, or by a pattern — into columns you name. |
| Merge columns | Join several columns into one with a separator you choose. |
| Rename / Reorder / Drop columns | Exactly what they say. |
| Fill blanks | With a constant, or by carrying the value above down. |
| Standardize categories | Map the twelve spellings of one value onto the one you want. |
| Enrich | Fill in a normalized value from a bundled reference table (below). |
| Sample | Take a number of rows or a percentage, optionally spread evenly across a column. |
| Anonymize | See below. |
When a value will not convert, DataTray does not quietly turn it into a blank: the row is
collected in … - rejects.csv beside the output, with the line number, the column, the original
text and the engine's own reason.
Enrichment is reference data, not intelligence
The Enrich operations — split a full name into parts, normalize a country, normalize a US state, check a postcode's shape, format a phone number as E.164, check an email's syntax, normalize a currency code, convert units — are lookups against small, versioned CSV tables that ship inside the app, with their source, licence and retrieval date recorded. They are data. They are not AI, they are not a service, and they do not learn anything about your file.
Finding personal data
DataTray automatically finds columns that look like personal data — email addresses, phone numbers, national identification numbers, payment card numbers, IBANs, IP addresses, dates of birth, postal addresses and person names — using shape and header rules, and reports each with a confidence.
The detection is a suggestion, and you confirm before anything is written. It is not guaranteed to be complete: a column DataTray did not flag may still contain personal data, and a column it flagged may not. A tool that silently decided what was personal would be worse than no tool, which is why nothing here is applied until you choose it.
The six anonymization actions — and what each one honestly is
This is the part to read carefully, because the words matter legally as well as practically. Only two of the six actions actually anonymise. The other four reduce or transform the data without making it anonymous, and DataTray prints the caveat beside each action in the panel and again in the report rather than letting you assume otherwise.
| Action | What it does | What it honestly is |
|---|---|---|
| Drop | The column is not written to the output at all. | The only action that removes the data rather than transforming it. |
| Redact | The value is destroyed. | The original cannot be recovered from the output. Rows may still be identifiable through the columns left untouched. |
| Mask | Keeps the first n, last n, or the email domain, and hides the rest. | Not anonymisation. The kept characters stay in the clear and the value keeps its original length, so a masked value is often re-identifiable from the rest of the row. |
| Hash | Replaces the value with a SHA-256 digest, optionally salted. | Pseudonymisation, not anonymisation. The same input always produces the same digest, so a value from a small or guessable set — an email address, a phone number, a customer id — can be recovered by hashing candidates until one matches. A salt stops a pre-computed lookup table and separates one run from another, but it is printed in the report, so it is not a secret. |
| Tokenize | Replaces each distinct value with ID-000001, ID-000002, … | Pseudonyms, not anonymisation. Anyone holding the token map can reverse them, and the tokens are numbered in the sort order of the underlying values, which reveals their relative order. Tokens are stable within one run only. |
| Generalize | Reduces precision: a date to its year, a postcode to its prefix, an age to a band. | Lowers precision but does not guarantee anonymity. A rare combination of generalised values — a birth year with a postcode prefix, say — can still single someone out. No k-anonymity check is performed. |
Two practical consequences:
- Hash and tokenize are what make joins survive. If you need two anonymized extracts to still line up on customer id, hash it with the same salt in both runs — and read the file the same way both times, because the digest is of the value as it was read.
- An empty salt is refused, rather than silently producing an unsalted hash that the report would still describe as salted.
Every action except Drop rewrites the column in place — same name, same position — so the output is a drop-in replacement for the original file. Empty cells stay empty rather than becoming a row of bullets, because "this cell was blank" and "this cell was hidden" are different facts.
The report
Ask for a report and you get an HTML document recording: every operation that ran and how many rows it affected, the columns that were not modified, and — for an anonymization run — each column, the action applied, the salt if one was used, and the honest caveat for that action quoted above. It is written for someone who has to review what was done, not for you to admire.
Output
Saved to the Cleaned subfolder of your output folder, with a rejects file and a report beside it
where they apply.
If something goes wrong here
- "Add at least one operation to the stack before running." The stack is empty. Pick something from the findings list, or add an operation by hand.
- My Excel file was refused. This tile reads delimited text, Parquet and JSON in this release. Convert the workbook with Data Converter first.
- Rows are missing and there is a rejects file. A conversion refused those values. Open
… - rejects.csv— it names the line, the column and the reason. - The rows came back in a different order. A tokenize operation groups values in order to number them. DataTray restores the original row order, but if you ever see a reordered result, say so in Feedback — that is a bug, not a design.
- The sample option I want is greyed out. Taking an exact number of rows and spreading the sample statistically are mutually exclusive; pick a percentage if you want the statistical method.
Clean & Anonymize — see what it does
Trailing whitespace, dates in four formats, numbers stored as text, the leading zeros that vanished, encoding that turned é into é. DataTray finds them without being asked and fixes them in one previewable stack of operations — and in the same pass it finds the personal data, so you can mask, hash or tokenize it before the file goes anywhere. Every finding is a deterministic rule over the column, explainable in one sentence. No model, no API key, nothing uploaded. Learn more about the Clean & Anonymize.
Windows 10/11 (x64) · Free · No account
Haven’t installed DataTray yet?
Free for Windows 10 and 11. No account, and nothing uploads.
Windows 10/11 (x64) · Free · No account