Scoring

Every completed verification receives a composite score (0–100) that drives the automated decision. The score combines identity verification results with compliance screening.

Score formula

The overall score blends two composites — the identity score (idvScore) and the compliance score (complianceScore):

overall = (idvScore × 0.6) + (complianceScore × 0.4)

The idvScore is the weighted average of the identity components below (each 0–100):

idvScore = weighted average of:
    faceMatch        (weight 0.25)
    liveness         (weight 0.25)
    documentQuality  (weight 0.10)
    documentAuth     (weight 0.10)
    nameMatch        (weight 0.10)
    dataConsistency  (weight 0.10)

Component	Weight	Source
faceMatch	0.25	FaceNet embedding similarity
liveness	0.25	Anti-spoofing analysis
documentQuality	0.10	Image quality assessment
documentAuth	0.10	5-layer document authenticity analysis
nameMatch	0.10	Name matching across sources
dataConsistency	0.10	Cross-reference of extracted data

The complianceScore (0–100, where 100 is clean) comes from sanctions / PEP / adverse-media screening — see Compliance screening below.

note

When a component doesn't apply to a verification — for example nameMatch when no expected name was provided, or mrzValidity for a non-MRZ document — it is excluded and the remaining component weights are renormalized, so idvScore always spans 0–100.

Score components

The score is derived from multiple sub-scores:

Document quality (documentQuality)

Image quality assessment — resolution, blur, glare, cropping. A low score here means the image is difficult to analyze reliably.

Document authenticity (documentAuth)

5-layer authenticity analysis:

Layer	Check	What it detects
D1	Template Matching	Document layout vs known templates
D2	Font Consistency	Font uniformity across text fields
D3	Photo Integrity	Photo replacement or manipulation
D4	Compression Analysis	Re-compression artifacts from editing
D5	Edge Consistency	Cut-and-paste or border manipulation

Face match (faceMatch)

Similarity between the selfie and the document photo, computed using FaceNet 512-dimensional embeddings.

• faceMatch — Overall match score (0–100)
• faceEmbeddingSimilarity — Raw cosine similarity of face embeddings
• faceMatchConfidence — Confidence level of the match

Liveness (liveness)

Anti-spoofing analysis with 5 layers:

Layer	Score	What it detects
Texture	textureScore	Printed photos, screen displays
Frequency	frequencyScore	Digital manipulation artifacts
Challenge	challengeScore	Correct completion of active challenges
Temporal	temporalScore	Frame-by-frame consistency
Quality	qualityScore	Image quality of liveness frames

MRZ validity (mrzValidity)

For ICAO 9303 documents (passports, some national IDs) — validates the Machine Readable Zone check digits and data consistency.

Data consistency (dataConsistency)

Cross-checks extracted document data (name, DOB, document number) against MRZ data and any government database results.

Name match (nameMatch)

Compares the name extracted from the document (OCR and MRZ) against the name provided at verification creation, using fuzzy matching algorithms.

Compliance screening

In addition to the verification score, compliance screening results from CEngine are factored into the decision logic (see below). The compliance screening includes:

• Sanctions screening — Checked against 9 international sanctions lists (65K+ entries)
• PEP screening — Checked against 764K+ Politically Exposed Person entries
• Adverse media — NLP analysis of news and media mentions

Compliance risk bands

Risk Band	Score Range	Meaning
LOW	80–100	No hits, clean screening
MEDIUM	50–79	Minor hits requiring review
HIGH	20–49	Significant hits
CRITICAL	0–19	Sanctions match or critical risk

Decision logic

The overall score drives an automated decision (compliance hard-stops are applied first, then the score bands):

Condition	Decision	Action
Sanctions hit	auto_reject	Immediately rejected regardless of score
CRITICAL or HIGH compliance risk	auto_reject	Rejected due to compliance risk
PEP hit (no sanctions)	manual_review	Queued for human review
MEDIUM compliance risk	manual_review	Queued for human review
overall >= 80	auto_approve	Automatically approved
overall 50–79	manual_review	Borderline, queued for review
overall < 50	auto_reject	Automatically rejected

note

Decision thresholds are configurable per tenant. Contact your account manager to adjust the auto-approve and auto-reject thresholds for your use case.

Score response

The complete score breakdown is included in the verification response and webhook payload:

{
  "scores": {
    "documentQuality": 95.0,
    "documentAuth": 97.5,
    "faceMatch": 98.1,
    "faceEmbeddingSimilarity": 0.94,
    "faceMatchConfidence": 0.97,
    "liveness": 94.5,
    "antiSpoofScores": {
      "textureScore": 96.0,
      "frequencyScore": 93.0,
      "challengeScore": 95.0,
      "temporalScore": 92.0,
      "qualityScore": 96.5,
      "spoofDetected": false
    },
    "docAuthScores": {
      "templateScore": 98.0,
      "fontScore": 97.0,
      "photoIntegrity": 96.5,
      "compressionScore": 98.0,
      "edgeScore": 97.5,
      "tamperDetected": false
    },
    "mrzValidity": 100.0,
    "dataConsistency": 95.0,
    "idvScore": 96.3,
    "complianceScore": 95.0,
    "complianceRiskBand": "LOW",
    "overall": 95.8
  }
}