BorderRev · Model 5 of 5

Every signal.
One score.
One decision.

The Risk Scoring model fuses the verdicts of all four BorderRev models into a single 0–100 risk score — with a signal-by-signal breakdown that tells an officer exactly why the flag fired.

score · req_9f2a44b HIGH RISK · 87
HS Classification incorrect
Valuation −62% below corridor
Origin transshipment flag
Transfer pricing related-party detected
Trader history 2 prior violations
Score 87 / 100
Three independent models fired on this declaration. Classification and valuation anomalies compound a known transshipment corridor. Routed to Standard Examination.
5
Signal sources fused per score
<1s
Score generation time
100%
Scores with explainable rationale
Examination efficiency vs manual triage

A declaration that passes one check can still be high risk.

Customs systems that check signals independently miss the compounding effect. A 10% valuation anomaly alone might be within tolerance. The same anomaly on a misclassified declaration from a flagged trader on a known transshipment route is a different story entirely.

Manual triage of multiple model outputs is slow, inconsistent, and cannot scale to national volumes. Officers looking at four separate verdicts in four separate systems are not seeing the full picture — they are seeing fragments.

The Risk Scoring model was built to solve exactly this. It holds all signals simultaneously, weights them against each other, and produces a single number that reflects the true compound risk of the declaration.

1 in 12
high-risk declarations correctly identified when signals are assessed in isolation
1 in 3
identified when all signals are fused — a 4× improvement
Detection rates from BorderHQ analysis across multiple national administrations using historical examination outcomes as ground truth.

From four verdicts to one decision.

The Score Engine ingests structured verdicts from each BorderRev model, overlays trader profile and external intelligence signals, and produces a composite risk score in under a second.

Model Signals

Structured verdicts from Classification, Valuation, Origin, and Transfer Pricing. Each verdict carries a confidence level and severity rating that the score engine uses to weight the signal.

Trader Profile

Compliance history, prior violations, declaration frequency, and sector risk profile. A first-time importer and a licensed trader with 1,200 clean declarations represent very different risk contexts.

External Intelligence

Sanctions hits, watchlist matches, commodity-level intelligence, and anonymised peer-country signals. Intelligence feeds update continuously from BorderHQ's managed data sources.

Score + Routing

A 0–100 composite score with a per-signal breakdown. The score maps to an examination queue automatically — no manual triage required.

Every number has a reason.

score · req_9f2a44b — 2026-05-26
Declaration Declarant : Fenix Trading Ltd (trader_id: TRD-00441) HS Code : 2933.49.90 (declared) Origin : CN → transit: AE → destination: national admin Value : USD 18,400 (declared) Signals Classification : incorrect severity: HIGH confidence: high Valuation : −62% corridor severity: HIGH confidence: high Origin : transshipment severity: MEDIUM confidence: medium Transfer pricing : related-party severity: MEDIUM confidence: medium Trader history : 2 violations severity: MEDIUM confidence: high Score Composite : 87 / 100 Band : HIGH RISK (80–97.9) Routing : Standard Examination Rationale Three independent models fired on this declaration. Classification (HIGH) and valuation (HIGH) anomalies are correlated — both point to under-invoicing of a misclassified pharmaceutical intermediate. Origin transshipment through AE is a known corridor for this commodity group. Transfer pricing flag and 2 prior violations amplify the base signal. Compound weighting applied: correlated classification+valuation pair treated as a single HIGH-severity signal with boosted weight. Actions Queue : EXAM-2026-4471 Priority : immediate Evidence package : generated → 4 model verdicts + trader profile appended Audit log : score_log · entry 2026-05-26T09:14:32Z · immutable

Every score routes itself.

The four examination bands map directly to workflow queues. No analyst needs to read four model outputs and make a judgement call — the score does it.

98–100
Priority Examination

Immediate physical inspection

Evidence package generated automatically from all four model verdicts and trader profile. Routed to examination queue with officer assignment.

80–97.9
Standard Examination

Documentary review

Risk-informed inspection criteria generated from the signal breakdown. Physical inspection criteria provided if documentary review escalates.

70–79.9
Post-clearance Audit

Released with audit flag

Declaration clears immediately. Audit flag retained for review within 30 days. Score and signal breakdown preserved in the audit record.

0–69.9
Auto-clear

No action required

Declaration passes. Score cached and written back to trader profile to update the baseline. Clean history reduces future scores for this trader.

01 · EXPLAINABLE BY DESIGN

A score without an explanation is an instruction without a reason.

Every score comes with a per-signal breakdown: which model fired, at what confidence, and how each signal contributed to the composite. Officers trust scores they can interrogate.

When a trader disputes a flag or an examiner wants to understand the basis of a queue assignment, the breakdown is already there. There is no secondary analysis step and no need to pull four separate model outputs.

Per-signal contribution Confidence levels Severity ratings Compound logic explained
Signal contribution — Score 87 / 100 87 Classification · 26 pts Valuation · 24 pts Origin · 14 pts Transfer pricing · 12 pts Trader + Intel · 11 pts 100 Compound boost applied: Classification + Valuation correlated pair treated as single high-severity signal with amplified weight
02 · COMPOUNDING LOGIC

The score engine is not additive.

Two medium signals on correlated dimensions — valuation and transfer pricing both firing on the same related-party group — produce a higher score than their sum. The model detects when signals reinforce each other.

This is the key difference between risk scoring and risk flagging. A flag system tells you how many things went wrong. The score engine tells you how wrong they are together.

Correlation detection Signal amplification Compound weighting
Risk compounding matrix Valuation anomaly → Origin flag → Low High Low High 18 Low risk 42 Medium risk 48 Medium risk 87 Score amplified Correlated risk sum would be 60 compound: 87
03 · TRADER PROFILE LEARNING

Clean history is evidence, not an assumption.

Every examination outcome feeds back into the trader's profile. A clean examination reduces future risk scores for that trader. A violation increases them. The system learns which traders are genuinely low-risk and reduces examination overhead over time.

This is not a static whitelist. A trader's score reduction is earned through confirmed clean declarations, not granted by administrative category. And it can be reversed if patterns change.

Outcome feedback loop Dynamic trader profiles Examination overhead reduction
80 50 20 3 clean exams → score reduction high risk Month 1 Month 3 Month 6 Month 9
04 · SOVEREIGN AND AUDITABLE

No declaration data needs to leave your control.

All scoring runs on-premises. No declaration data, no trader profiles, and no examination records leave the administration's infrastructure. Every score is retained in an immutable audit log.

The audit log is append-only and tamper-evident. Every score entry records the input signals, the weights applied, the composite result, and the routing outcome. An administration can reconstruct any score decision from the log alone.

On-premises deployment Immutable audit log No data egress Air-gapped mode
Administration boundary Score Engine on-premises Trader DB on-premises Audit Log immutable · on-premises Intel Feed managed · on-premises No data egress all signals processed and retained inside boundary

Ready to see what your declaration data actually looks like through all five lenses?

We offer a short, no-risk trial against a sample of your historical declarations — we return cited findings and composite risk scores across all five BorderRev models. No long engagement, no procurement overhead.

Request a briefing Back to BorderRev