What's new in PII Anomalyzer
Every release, newest first.
1.14.2
LatestA confidence cutoff for each type, far fewer false findings to review, private file names on export, and graphics card support on Windows.
New features
Set a confidence cutoff for each type
Every kind of information the AI finds now has its own confidence cutoff, so you can make one type stricter or looser without changing the rest. Point at a type in the sidebar and click its number to set its cutoff, or click Use main slider to put it back. Types without their own cutoff follow the main slider, now called Confidence Cutoff (default 0.55). Because the AI often mistakes legal role words such as "the Plaintiff" for names, Person starts at 0.90 and Organization at 0.80. Cutoffs apply only to the AI's scored findings, never to format matches, your lists or your Custom Patterns. If you had lowered the slider to catch more names, lower Person and Organization too.
Numbered file names on export
A de-identified file can still give a person away by its name, such as John_Smith_Contract.deid.pdf. Export now has a File Names choice: Original name (the default) or Numbered (document_001.deid.pdf). The app keeps a private record of which file became which number, so keep it as safe as a restore key. If you keep hidden data and metadata, a PDF's title can still hold the original name.
Uses your graphics card on Windows
On Windows, PII Anomalyzer can now use a supported graphics card to read scans and find personal information. Automatic (the default) uses a supported dedicated card when there is one, and the processor otherwise. The new Settings → Performance page shows what Automatic picked and lets you choose GPU or CPU for each task. Your documents stay on your computer either way. Read several pages at once has moved to this page as Read scans faster when memory allows, and your earlier choice is kept.
Improvements
Far fewer false findings
Common words that were flagged once are no longer copied through the whole document, and overlapping findings of the same kind now become one. A name found once no longer lands on ordinary words with the same letters, such as "will" for Will or "rose" for Rose. Tax words such as "IRS" or "backup withholding" are much less often flagged as tax ID numbers. Long legal documents show far fewer rows to review.
See how each item was found
In the Results table, the Method column is now Found by: AI, Format check, Your rule, Matched, Form field, or Your edit. Rows without an AI score show Not scored under Confidence and sort last. Point at a cell to see what it means. CSV and Excel exports use the same names.
Transcripts that store every word twice
Some PDFs, including some court transcripts, store each word twice at the same spot: once visible and once hidden. The app now reads such a word once, so the text and results no longer show every word doubled. Redaction still covers both copies.
Merge names across documents
When you import several files together, the Identifier dropdown now lists the identifiers from all of them, so Jane in one file can be joined to Jane Roe in another. These merges are saved in Manual Merges, where you can undo them.
Bug fixes
- The Confidence Cutoff now filters only the AI's scored findings, so Social Security and phone numbers stay found at any setting. With a high saved cutoff, more dates and places may appear; untick a type under Entity Selection to hide it.
- A phone number written with a narrow or thin space is now found and covered in PDFs and Office documents. Before, it could stay readable in a Redact export.
- A Social Security number right before a street address in a table is found as a Social Security number, not as part of the address.
- A number in Social Security number format is labeled as one, even when the AI guesses it's a phone number.
- Highlight exports no longer include a text file. Plain text can't show highlights, and the file held the original text.
- Clicking a Help search result opens the page at the first match, with every match highlighted.
- A name that wraps onto a line starting with a word and a colon, such as "Counsel present: Robert" then "Smith: for the plaintiff.", is now covered in full. Before, the surname could stay readable. If a find runs into a label on the next line, the label is now covered too.
1.14.0
Tell the app exactly what to hide, find settings in one place, and get better results from scanned documents.
New features
Tell the app exactly what to hide
Deny List entries can now set a term's category and its replacement, so Jane Roe always becomes <<CLIENT>>. Custom Patterns use regular expressions (regex) to find formats only your practice uses, such as case numbers, and Test previews the matches before you add a pattern. Import either list from CSV or Excel, or turn off automatic detection and run only your own lists and patterns.
All your settings in one place
Placeholder style, delimiters, Restoration Mode, theme and font size now live in one window under File → Settings, and the app remembers your choices between sessions.
Restore text from anywhere in your session
Paste or load text you de-identified earlier in the same session, even after a reviewer or an AI tool has worked on it, and Re-Identify puts the original values back. It no longer needs the file to match a document you have open, and short placeholders such as <<PER2>> are restored too.
Canadian identifiers
Social Insurance Numbers, Business Numbers and Canadian bank account details are now detected, and Canadian addresses are recognized in both English and French layouts. The Brazil CPF and CNPJ types have been retired.
Improvements
Better results from scanned documents
Sideways, upside-down and crooked pages are straightened before they're read, and two-column pages and tables are read in the right order. PDFs that are mostly scans are handled the thorough way automatically, and text read from a scan keeps its lines and paragraphs, so results are easier to review.
More identifiers caught beside their labels
Employer ID numbers (EIN), bank routing numbers, SWIFT/BIC codes and insurance policy numbers are now detected next to their labels, and a full City, ST 12345 line is covered, ZIP code included.
One simple choice for reading documents
How should documents be read? is now a single dropdown under Import Settings. Its most thorough option, which reads every page as an image, now works on Word and Excel files as well as PDFs.
Faster scans
On computers with plenty of free memory, the app now reads two scanned pages at a time, so long scans finish sooner. You can turn this off under Settings → Import → Read several pages at once.
Stays responsive with large files
Large PDFs, exports and result updates run in the background, so you can keep working, and exports show a progress bar.
Tidier Replace documents
In a Replace PDF, each placeholder is now sized to about 80% of the text it replaces, so the page keeps its layout. A placeholder too wide for its line sits just above it. You can adjust the sizing under Replace Overflow in Settings.
A cleaner review list
The default detection threshold is higher, so fewer low-confidence matches reach your review list. Lower the Detection Threshold slider if you'd rather see them.
Consistent placeholder numbers
Placeholder numbers now stay the same for your whole session, even when you add or remove files. They restart only when you choose De-Identify All or Clear All, and the status line tells you when that happens.
Fewer unnecessary warnings
When you choose Stored text only, the app no longer warns you about the scanned pages you asked it to skip. You get one short note at import instead.
Bug fixes
- On scanned pages, black boxes, placeholders and highlights now land on the printed words instead of beside them.
- An email address split across two lines is detected as one address.
- A name found once is also covered where it ends a sentence or continues onto the next line.
- When two detections overlap, all the text flagged by either one is covered. This can also cover a few nearby words that aren't sensitive.
- De-Identify no longer stalls after you edit text in a long document.
- A file that can't be opened, for example because it's damaged or open in another program, now shows a clear message instead of a file path.
1.13.0
Import photos directly, keep your lists between sessions, and get clearer messages.
New features
Import photos and scanned images
Add a JPG, PNG or TIFF file and it's read like a scanned page. Phone photos are turned the right way up, and a multi-page fax arrives as separate pages.
Your lists are saved
Your Allow and Deny Lists are kept when you close the app, and the Excel results export includes both on their own sheets, so you have a record of what was held back and what was always removed.
Photo location data removed
Photos can carry hidden camera details, often including where they were taken. Redact, Replace and Mask now remove this data from photos in the output.
Improvements
Consistent short codes
Every de-identified PDF now uses the same short code for the same kind of information, in every document and every batch. Most codes changed from earlier versions. Files you've already produced aren't affected, but update any scripts, templates or saved searches that look for the old codes.
Clear answers when something is skipped
If a run can't fully check a document, the app tells you why and where to look. Advice appears once rather than once per file, and exports include only documents that were successfully de-identified.
Safer restoring
An imported file is restored only against the document it actually came from. If there's no match, the app says so.
Protection against lost work
De-identified files are kept in a private folder inside the app's own data area while you work, so a crash or a large batch can't lose them.
Small things that add up
Select All now selects only the rows your filter shows, and your Include abbreviation legend choice is remembered between sessions.
Short codes that changed in 1.13
The codes you're most likely to have seen. Person (PER), Organization
(ORG), Location (LOC), Date and time (DAT) and
ZIP code (ZIP) did not change.
| What it marks | Old code | New code |
|---|---|---|
| Address | ADD | ADR |
| City | CIT | CTY |
| Company | COM | CMP |
| Country | COU | CNY |
| Credit card number | CRE | CCN |
| Email address | EMA | EML |
| First name | FIR | FNM |
| Health insurance ID | HEA | HIN |
| ID card | ID_ | IDC |
| IP address | IP_ | IPA |
| Last name | LAS | LNM |
| License plate | LIC | LPL |
| Other personal information | GEN | PII |
| Phone number | PHO | PHN |
| Postal code | POS | PSC |
| Social Security number | US_ | SSN |
| State or province | STA | STP |
| Title | TIT | TTL |
| Username | USE | USR |
Bug fixes
- Coverage warnings no longer flag table borders and shading.
- Redaction boxes line up exactly with drawn logos, including at the page edge.
- An image stored in a file but not shown on any page now gets a clear explanation instead of an error.