// VERTEX AGILITY · AI HEALTH SCAN

Loan Default Risk v3

Scan date 2026-08-07 · reference 8,000 rows vs current 6,500 rows

Overall status: RED

Headline

2 of 7 features show material drift, led by 'region' (PSI 1.54).
The model's own output distribution has shifted (PSI 2.03); average score moved 0.25 to 0.41. This is an early warning even before labels arrive.
Data-quality: 'employment_years' missingness rose 0% to 7%.
Governance readiness is 2/5; the evidence a risk function would ask for is incomplete.

Scorecard

Data quality

RED

2 issue(s)

Input drift

RED

2 red / 0 amber

Output drift

RED

PSI 2.03

Governance

RED

2/5 controls

Input drift by feature

Feature	Type	PSI	KS	Wass./JS	Status
region	categorical	1.536	-	0.047	RED
dti_ratio	numeric	0.298	0.22	0.426	RED
annual_income	numeric	0.100	0.13	0.198	GREEN
loan_amount	numeric	0.085	0.12	0.301	GREEN
channel	categorical	0.070	-	0.009	GREEN
age	numeric	0.040	0.09	0.141	GREEN
employment_years	numeric	0.003	0.01	0.009	GREEN

PSI convention: <0.10 stable, 0.10–0.25 moderate, >0.25 material. KS for numeric features; JS divergence shown for categoricals.

Data quality

● 'employment_years' missingness rose 0% to 7%.
● 'region': unseen category ['NE-online'] (12.9% of rows).

Distribution shift

Governance readiness RED

✓ Predictions and features are logged in production
✓ A reference / training baseline is retained
✗ A model card or documentation exists
✗ Evaluation history is kept
✗ Automated alerting is wired up

What this means and what we recommend

Material issues are present that are unlikely to be visible on your current dashboards. We recommend a full Assurance Audit to instrument monitoring properly, confirm whether the drift is affecting decisions, and set an SLA baseline.

Automated scan across a single model and a fixed reference window. It is a first-look health indicator, not a full assurance review: it does not confirm root cause, cover multiple models, or instrument standing monitoring. Those are the scope of a VA Assurance Audit. Findings are indicative and should be read with a VA engineer.