Public engine documentation
Email Signal Explorer
Gorganizer's scoring engine emits 2,119 unique signals across six modules — headers, sender, subject, body, attachments, and structural metadata. Browse by category below; each category page lists its full signal set, and every signal links to a full explanation, severity tier, and false-positive notes.
Protective
Prevents deletion. Always wins over trash signals.
Threat
Phishing, BEC, scam — strong contributor to the trash decision.
Warning
Bulk / marketing / mild spam. Insufficient on its own.
Learned
Per-user history boost. Only fires for senders you have trashed.
Protective signals
373 signals
- "Accepted" (protected)
- Account suspended / access revoked (protected)
- Account activity / statement summary (protected)
- Account balance warning (protected)
- Account change notification (protected)
Phishing & impersonation
607 signals
- Fake plan administrator or financial advisor claiming a COVID or hardship early withdrawal from a 401k is available and that the target can avoid the 10% IRS penalty by submitting a claim within 30 days — advance-fee or credential-harvest fraud; real 401k hardship withdrawals are initiated through authenticated plan administrator portals, never cold email with penalty-avoidance claims.
- Fake ACA marketplace or health insurance administrator claiming coverage will be cancelled unless income or enrollment is verified via email link — credential-harvest attack; real ACA coverage changes are communicated through healthcare.gov or state marketplace portals, never cold email links demanding income verification to prevent cancellation.
- Fake agentic-AI permission-grant harvest — approve/authorize/delegate + inbox/calendar/repo scope + off-platform OAuth link.
- Fake AI API runaway-spend / overage alert (OpenAI/Anthropic/Claude API) to harvest API keys or cloud credentials from non-official sender
- AI Deepfake Extortion Threat
Scams & fraud
347 signals
- Advance-fee personal loan scam
- Advance-fee prize release demand
- AI voice-clone charity donation scam — deepfake audio/video of a celebrity or executive soliciting urgent crypto/wire donation at a non-official URL. FBI IC3 2025; Chainalysis 2026 charity-scam spike post-disaster.
- Background check / data removal scam — your records publicly exposed + pay to remove or enter personal info
- Business grant / govt funding scam — free SBA/government grant + upfront processing fee to release funds
Sender / domain
22 signals
- AI-generated invoice BEC with lookalike domain — plausible invoice from a typosquat vendor domain with changed banking / wire-transfer details mid-transaction. Abnormal Security 2025-2026; FBI IC3 2025 BEC $2.9B; distinct from fake-invoice-callback (phone-number shape).
- Automated sender address (notifications@, alerts@, etc.)
- BIMI logo checkmark on lookalike phishing domain (brand impersonation via VMC certificate)
- Deep subdomain (click-tracking infrastructure)
- Display name is an email address (impersonation)
Subject line
113 signals
- "Action required" / "Complete your profile"
- "Alert:" prefix subject
- Subject is all-caps
- "Archived" subject
- "Awaiting your response" subject
Body content
65 signals
- Account phishing — suspension threat + action demand
- Advance-fee fraud (419) — identity claim + fund transfer request
- Password-protected archive with password disclosed in body — malware delivery
- Blob/JavaScript URI in Anchor Href
- Body mentions an attachment but the email has none — phishing primer for the next link click
Email headers
44 signals
- Auto-generated email
- Brand Domain, DKIM Absent
- Brand Domain, DKIM Verified
- Read-Receipt Harvesting
- DKIM signed by different domain
Attachments
7 signals
- Archive attachment (.zip/.rar/.7z — common spam delivery)
- Attachment filename mixes Latin with Cyrillic/Greek (homograph)
- Attachment has no filename — malformed bulk mailer or deliberate content-scanner evasion
- Bidi override in attachment filename — invoice.exe masquerading as invoice.pdf
- PDF + credential harvest subject
Marketing & bulk mail
135 signals
- Abandoned cart reminder (automated e-commerce re-engagement)
- Adoption process notification (protected)
- Year in review / annual summary / Wrapped
- API rate limit / quota notification
- App update available notification
Structural / Gmail metadata
21 signals
- Declared text/plain but contains HTML (filter evasion)
- Date header set far in the past (bottom-of-inbox hide)
- Date header set far in the future (top-of-inbox manipulation)
- In Gmail Forums tab
- Marked important by Gmail
Other
385 signals
- Account inactivity / win-back
- Fake acquired-vendor rebrand / change-of-accounts (CoA) lure — "Vendor X is now Vendor Y, please update bank details on file." Mimics legit M&A churn; vendor-name change is the tell. Real vendor M&A bank-detail changes flow through the AP-system change-notification process with verbal verification through a known phone contact, never via a single inbound email demanding wire-redirect on a deadline. Sender NOT on the merchant / bank canonical-allowlist (stripe.com, paypal.com, amazon.com, apple.com, visa.com, mastercard.com, americanexpress.com, discover.com, klarna.com, adyen.com, square.com, squareup.com, shopify.com, wise.com, revolut.com, jpmorgan.com, chase.com, bankofamerica.com, wellsfargo.com, citi.com, hsbc.com, barclays.com, deutsche-bank.com, bnpparibas.com, ing.com, santander.com, rabobank.com, nordea.com, seb.se, swedbank.com, handelsbanken.com). Distinct from R6/R7/R8 generic vendor / merchant spoofs — this signal is specifically the M&A-rebrand bank-detail-redirect variant, an AP-fraud / wire-redirect precursor that bypasses FP-control on standard merchant-spoof signals because the framing is "we changed banks because of acquisition" rather than "your payment failed." Source: Red-Team R8 multi-agent council S2 (social-engineering specialist).
- Fake Adobe Sign / Acrobat Sign document-signing request from non-adobe.com sender
- Spoofed notification claiming an AI assistant / agentic tool needs expanded OAuth/tool permissions. "Approve expanded permissions within 24 hours: grant access to your calendar, email, and file storage." Targets users of AI assistants (ChatGPT, Copilot, Claude, Gemini). Sender NOT a canonical AI vendor (anthropic.com, openai.com, google.com, microsoft.com, etc.). Label-only: engine flags but cannot patch the agent platform — action is always label, never trash. Source: Red-Team R8 multi-agent council C5 (agentic-AI specialist).
- Agentic-AI vishing lure — email directs victim to call a phone number where an AI voice agent conducts the credential-harvest or fraud
Want to see them in action?
Connect your Gmail in 10 seconds and Gorganizer will show you exactly which signals fired on every email — colour-coded by severity, with full explanations.
Get started