PII & Privacy Transparency | Montana Blotter

Posts Audited

512

Clean

497

Manual Review

Pending

% With Flags

63.1%

323 flagged posts

PII types detected

Type	Detections	Why we redact it
`home_address`	2282	Street addresses — redaction triggered when a residential identifier (apartment, trailer, home) is paired with a personal label.
`dob`	57	Dates of birth labeled as personal (DOB, birthday, born on). Birthdays that appear as event context are kept.
`other`	2	Other PII surfaced by Claude during tone/context review. Re-reviewed manually before publication.
`phone`	2	Phone numbers — redaction triggered when the number is paired with a personal label (caller, victim, witness).

Severity breakdown

High = direct identifier (SSN, MT DL). Stops publication automatically. Medium = contextual identifier (residential address, personal phone). Redacted and editor-reviewed. Low = contextual date or label (DOB, age). Reviewed before publication.

Medium: 2267

High: 19

Low: 57

1. Regex scan

Every post draft is scanned against a list of pattern detectors — SSN shape, MT DL format, US phone, residential street address with a personal-context label, and labeled DOBs. The detectors are deliberately conservative: a street address without a residential context word (apartment, residence, home, trailer) is allowed because it is typically the public incident location.

2. Claude review

When the API is available, the post is also passed to a second-pass model that flags contextual PII the regex misses — for example, an indirect reference to "her boyfriend's truck, which is usually parked at 412 Spruce" — and returns a clean rewrite. The clean version is the one that publishes.

3. Editor sign-off

Posts flagged as manual_review are not published until an editor reviews them. A clean post is published automatically. A pending post has not yet been audited.

4. The corrections log

If something slips through, we log the fix in the public corrections log. Editorial framing changes, factual errors, and PII redactions after the fact are all visible there.

What we scan for, what we catch, what we redact.

How clean is the public post corpus?