Private Preview
Join Waitlist →

Smart Detection

How Redactorr finds sensitive data through multi-layer pattern matching and validation

5 min readAuthor: Redactorr Support Team · [email protected]Last reviewed: March 2026

Outcome

Decide whether a detection is correct, missed, or too broad, then choose the next fix.

Your progress

Mark each step as you complete it. This is saved only in this browser.

0 / 3 steps complete

Before you start

Have the document open in review mode.
Know whether the issue is a missed item, an extra mark, or a custom internal format.
Use fake examples when describing custom identifiers.

Visual frame

Review asset

Smart Detection storyboard

Synthetic detection frame only. It presents detection as a review aid and avoids promising perfect coverage.

SignalsPattern plus context
ConfidenceReview priority
DecisionUser confirms

Synthetic frame

Detection signals

A finding is more than a highlighted value.

FormatEmail-like
ContextContact field
Example keywordCheck carefully
Signals help prioritise review, but the user still decides.

Instructional detection review frame

A simplified review frame showing how to compare detections, ignored items, and custom-pattern paths.

DetectedItems Redactorr found and recommends reviewing.
Needs decisionMedium or uncertain items should be checked by the user.
Custom formatOrganisation-specific values may need custom patterns.

Guided steps

Follow the task, then check the result.

Name what happened

Choose whether Redactorr missed sensitive text, marked too much, used the wrong category, or needs a custom pattern for your organisation.

Detection list with item status and confidence labels.
Success check: You can describe the issue without sharing raw private text.

Fix the current review state

Confirm expected detections and ignore anything that should remain visible. If a whole internal format is missing, move to custom patterns rather than manually fixing every instance.

Confirm and ignore actions in the review panel.
Success check: The current output reflects your decision for each visible item.

Check the redacted output before trusting it

Read the final output like a recipient would. Make sure raw sensitive values are gone and harmless text is still readable.

Before and after review strip with redacted placeholders.
Success check: The final output has no obvious missed private values and no unnecessary redaction that would break the task.

Branch questions

Completion check

Missed, extra, and uncertain items have been reviewed.
Custom internal formats are routed to custom patterns with fake examples.
The exported or copied output still makes sense after redaction.

Support boundary

Support can use by default

  • Detection category labels and confidence band.
  • Selected symptom, article slug, and review step.
  • Synthetic examples the user writes deliberately.
  • Intent ID
  • Article slug
  • App route
  • Browser and viewport

Requires your consent

  • User-written description
  • Email address
  • Explicit attachment
  • Redacted sample
  • Support bundle previewed to the user

Not collected by default

  • Raw original document text
  • Full local file paths
  • Pasted private content
  • Unmasked screenshots
  • Replay capture

Safe support summary

Copy a scrubbed handoff.

Redactorr KB support summary

Article: /knowledge-base/smart-detection
Selected issue: A supported common value type is consistently missed.
Playbook progress: 0/3 steps marked complete

Safe context to include:
- Detection category labels and confidence band.
- Selected symptom, article slug, and review step.
- Synthetic examples the user writes deliberately.
- Intent ID
- Article slug
- App route
- Browser and viewport

Requires explicit consent:
- User-written description
- Email address
- Explicit attachment
- Redacted sample
- Support bundle previewed to the user

Do not include by default:
- Raw original document text
- Full local file paths
- Pasted private content
- Unmasked screenshots
- Replay capture

User note:
- Describe the step and symptom without pasting raw document text, secrets, files, or restoration material.

Article details

Smart Detection: Finding Needles in Haystacks

Redactorr does not just look for obvious patterns like "[email protected]". It runs hundreds of patterns with format verification and context analysis to find sensitive data accurately.

What Gets Detected

Personal identifiers

  • Email addresses: All formats, including plus addressing and subdomains
  • Phone numbers: Australian mobile and landline formats, international numbers
  • Tax File Numbers, Medicare numbers, ABNs, ACNs, BSBs
  • Addresses: Street addresses, PO boxes
  • Names: First, last, full names with context analysis

Credentials and secrets

  • API keys: 50+ providers (AWS, Stripe, GitHub, and others)
  • OAuth tokens, access tokens, refresh tokens
  • Passwords in code, configuration files, connection strings
  • Private keys: RSA, SSH, PGP

Financial data

  • Credit cards: Visa, Mastercard, Amex, Discover (format verified)
  • Bank accounts: BSBs, account numbers
  • IBAN: International bank account numbers
  • Crypto: Wallet addresses

Healthcare data

  • Medical record numbers
  • Patient identifiers, health fund member IDs
  • AHPRA registration numbers

Custom patterns

  • Your organisation's ID formats
  • Internal project codes
  • Customer account numbers

How It Works

Pattern Screening**

1 / 1

Redactorr scans your text with hundreds of detection patterns, each tuned for a specific type of sensitive data.

Validation**

1 / 1

Every match is verified:

  • Credit cards are checked against standard format rules
  • Emails must have valid domain structure
  • Account numbers must follow known structural rules
  • API keys must match provider-specific formats

Context Analysis**

1 / 1

The surrounding text matters:

  • Is this in a comment or documentation? (Probably an example)
  • Is it near keywords like "fake", "test", "example"? (Probably not real)
  • Is it in a code snippet? (Might be a placeholder)

Confidence Scoring**

1 / 1

Each detection gets a confidence score:

  • High: Very likely sensitive — flagged immediately
  • Medium: Probably sensitive — shown for review
  • Low: Possibly sensitive — held back in standard mode

Real-World Examples

Email detection Catches:

Skips:

API key detection Catches:

  • sk_sample_live_51Hxyz123456789abcdefghijklmn (Stripe)
  • AKIAIOSFODNN7EXAMPLE (AWS)
  • ghp_1234567890abcdefghijklmnopqrstuvwxyz (GitHub)

Phone number detection Catches:

  • 0412 345 678 (Australian mobile)
  • +61 2 9876 5432 (Australian landline)
  • (03) 9876 5432

Custom Patterns

Do not see your organisation's ID format? Add a custom pattern:

Go to Settings → Custom Patterns

Click "Add Pattern"

Name it (e.g., "Customer ID")

Provide the format pattern: CUST-d{6}

Test with sample data

Save and activate

1 / 6

Now Redactorr will detect "CUST-123456" as sensitive data.

Still stuck?

Copy a safe article handoff.

Start support with the article, the issue, and safe context. Raw document text, files, restoration material, and unredacted screenshots stay out unless you explicitly choose otherwise.

Support can start here
  • Article slug
  • Selected issue
  • Screen name
  • Action name
Redactorr support case

Source: Article playbook: smart-detection
Route: /knowledge-base/smart-detection
Selected issue: A supported common value type is consistently missed.

Safe context:
- Detection category labels and confidence band.
- Selected symptom, article slug, and review step.
- Synthetic examples the user writes deliberately.
- Article slug
- Selected issue
- Screen name or article section
- Action name
- Browser and viewport

Only include with explicit consent:
- User-written description
- Email address
- Explicit attachment
- Redacted sample
- Support bundle previewed to the user

Do not include by default:
- Raw original document text
- Full local file paths
- Pasted private content
- Unmasked screenshots
- Replay capture
- Hidden diagnostic uploads

User note:
- Describe what you tried and what happened. Do not paste raw document text, secrets, files, restoration material, or unredacted screenshots.

Support case builder

One case format, wherever you start.

This is the same support case shape used by diagnostics and article handoffs.

Safe to include
  • Detection category labels and confidence band.
  • Selected symptom, article slug, and review step.
  • Synthetic examples the user writes deliberately.
  • Article slug
  • Selected issue
  • Screen name or article section
Consent boundary

Private material belongs behind an explicit consent step, not in the initial case.

6 default exclusions

Before you copy0/4 ready
Open safe support guide