Defeating Perfect Signal Injection
Fraud rings inject manufactured telemetry past pattern-matching engines. NonaShield verifies physical hardware signals instead of trusting the JSON packet.
What the Backend Actually Checks.
A "Golden Payload" has to survive every stage below, in order, before it's trusted — not one pattern-match on a JSON blob. The same pipeline also carries 50+ real fraud-pattern vectors, from digital-arrest coercion signatures to SIM-swap and mule-account correlation.
Token & Header Validation
Parses the signed request token, enforces ±60s clock skew, ≥256-bit nonce entropy, expiry, and an allowlisted action string — before anything downstream is trusted.
Replay Rejection
In-process 300-second replay window, then a cross-instance nonce check against Redis so a captured request can't be replayed from a different edge node.
Signature Verification
ECDSA P-256-SHA256 signature over the canonical request, checked against that specific device's registered public key — not a shared secret.
Crypto Gate — 8-Element Trust Orchestration
Asserts public key registry lookup, hardware attestation, device-binding token, signature, nonce + timestamp, decision log entry, policy version, and orchestrator identity — all eight, fail-closed — before a verified request is ever released.
Certificate Revocation
Live OCSP check on every non-root certificate in the chain — fails hard on a definitive REVOKED response.
IP / ASN Reputation
Blocks Tor exit nodes, anonymous VPNs and hosting-provider ASNs against a maintained hostile-ASN table, cached 10 minutes at the edge.
Geo-Velocity & GPS-vs-IP Cross-Check
Haversine velocity between consecutive locations flags impossible travel at ≥900 km/h; device-reported GPS is separately cross-checked against edge-derived IP geolocation.
Bot & Automation Scoring
Composite 0–100 score from request-burst rate, User-Agent fingerprinting, and missing-SDK-header checks — fused downstream, never a standalone block.
Evidence Write
Only after every check above does the decision get hash-chained and dual-signed (device + server ECDSA) into the immutable evidence ledger.
Watch the Attack Fail.
The Verdict.
"Feedzai and Riskified are Pattern Matchers. NonaShield is a Truth Verifier."
What this case means for your app
Signal injection defeats checks that trust values the device reports. If an attacker can feed an app a plausible location, sensor reading or device profile, a purely statistical model sees a clean session.
The defence is to make the proof something the attacker cannot supply. A signature from a key held in secure hardware, verified separately at the edge and in the backend, cannot be forged by editing a value in memory. Cross-checking independent layers also exposes injected signals that contradict each other.
See hardware-backed possession and the zero-spoof architecture.
Frequently asked questions
What is signal injection?
Attackers feed an app or backend false but plausible device and sensor signals so a fraudulent session looks clean.
How does NonaShield defend against it?
Signals are bound to a hardware-held key and verified at the edge and in the backend, so injected values fail the signature or contradict other layers.
Why do perfect fake signals still fail?
Because the proof comes from a key that never leaves the device hardware, not from values the attacker can copy.