redactdeskOpen the tool

SUPPORT

Support

Limits, answers to common questions, and how to reach us.

Contact

Email redactdesk@outerscope.xyz. We reply as soon as we can.

Questions

What are the limits?

One file up to 1 MB and 5,000 records, with no record over 90 KB. Accepted formats are txt, csv, json, jsonl and pdf. A record is a paragraph line in txt, a data row in csv, an array element in json, a line in jsonl, and a paragraph block in a PDF. A PDF can be up to 10 MB and 200 pages, and the text read from it must stay within the same 1 MB, 5,000 record and 90 KB limits.

Which PDFs work, and what do I get back?

Only PDFs with a text layer, meaning text you can select. A scanned or image-only PDF is refused (there is no OCR), and so is a password-protected or damaged one. The cleaned PDF is a text-only rebuild: it keeps the text and the page breaks, not the original layout, images or fonts. Images, annotations, form fields and bookmarks are not read and are not in the output. The cleaned PDF uses a standard Latin font, so a character it cannot draw (Cyrillic, CJK, emoji and the like) is written as ? and the count is shown on the review screen; the cleaned text file keeps the real characters. In block mode a PDF drops every whole page that contains a detected entity. You also get the cleaned text and a manifest.

What do the three modes do?

Redact replaces each detected entity with its type in brackets, such as [EMAIL_ADDRESS]. Mask replaces each character of the entity with an asterisk and keeps its length. Block drops every record that contains a detected entity and counts it in the report; for a PDF it drops the whole page.

What is in the manifest?

The file name and format, record counts (scanned, changed, blocked), the mode, the selected types, per-type counts, the scan time and tool version, and a detection note. For a PDF it also holds the page counts, how many characters became ?, and a note that the cleaned PDF is a text-only rebuild. It holds no raw values and no hashes of values.

Why was a record missed?

Detection is automatic and imperfect. Short values with no surrounding text, such as a bare number, are the most likely to be missed. Spot-check the cleaned file before you sell it.

What if a scan fails?

The run stops at the batch that failed and keeps the counts so far. Use Retry from that batch, or start over. If the engine is warming up the first batch can take about 12 seconds.

Report a missed entity

Describe what kind of value was missed and the shape of the text around it. Do not send real personal data: replace the value with a made-up one of the same shape, for example a different phone number in the same sentence.