What is a Deterministic Layer?

A deterministic layer is based on fixed, rule-driven logic. If X happens, then Y is triggered. Think of it as a digital tripwire, designed to flag anything that violates a known constraint.

Examples of Deterministic Layers in Fortza:

• Address Validation: If the provided address fails verification against Google’s Address API, the build is blocked or escalated.
  
  
• Geolocation Hotspot: If a transaction occurs in a confirmed fraud hotspot, it’s flagged based on past abuse linked to that location.
  
  
• Historical Customer Data: If the behavior in this session deviates significantly from the user’s own established history, it's considered a deterministic flag.

Why We Use It:

Deterministic layers are fast, reliable, and built on known truths. They provide immediate certainty, ideal for catching repeat fraud patterns, rule violations, or identity manipulation. But they have limits: they can only catch what’s already been defined.

What is a Non-Deterministic Layer?

Non-deterministic layers use probabilistic models, AI, and behavioral heuristics to make judgment calls. They analyze intent, emotional state, and deviations from expected norms, not just hard-coded violations.

Examples of Non-Deterministic Layers in Fortza: 

• Encoder-Decoder AI: Detects subtle deviations in session flow and behavior, things that feel off, even if they don’t break a rule.
  
  
• Psychosocial Analysis (e.g., Name Recognition, Moral Drift): Interprets emotional tone, justification language, and social cues that may indicate manipulation or fraud rationalization.
  
  
• Anomaly Detection: Flags statistically unusual behavior that doesn’t match user or cohort norms, even if it’s never been seen before.
  
  
• FBI Identity Check: Uses partial-match logic to identify potential overlaps with known flagged identities, but lacks biometric or definitive verification, so it’s informative, not fatal.
  
  
• Domain Validation: Scores email domains based on age, DNS structure, MX records, and known burner lists, adding contextual risk but not certainty.
  
  
• Zip Code Proximity: Raises risk when a transaction occurs near previously reported fraud zones, but correlation doesn’t imply direct intent.
  
  

Why We Use It: 

Fraud doesn’t always look like fraud, especially in first-party fraud, social engineering, or insider threats. Non-deterministic layers allow Fortza to infer intent, detect emerging fraud tactics, and surface early behavioral indicators.

They give us nuance, adaptability, and early warning, critical tools in a landscape where today’s fraud often mimics normal behavior.

Here’s the punchline: Fraud isn’t either/or. It’s both/and.

The data shows us that most successful fraudsters today fall into two categories:

• Blatant actors who exploit known technical gaps (deterministic layers catch them fast).
  
  
• Sophisticated mimics who operate inside the lines, slowly escalating intent (non-deterministic layers are the only defense).
  
  

By using both types of layers, Fortza detects:

• First-time synthetic identities that sneak past traditional KYC.
  
  
• Behavioral mimicry, where a fraudster builds a fake profile over time.
  
  
• Impersonation patterns, like name overlaps or suspicious email syntax.
  
  
• Fraud clusters, where behavioral and geographic signals converge.
  
  

Fortza's Philosophy:

Redundancy Is Resilience

We don’t trust a single signal. We trust the convergence of signals.

That’s why Fortza’s stack doesn’t rely on just a rules engine, or just an AI model. It uses a conversation between deterministic logic, behavioral inference, and psychosocial insight. The result? A fraud signal that’s smarter, faster, and harder to fool.