Model Risk Governance Validator

# SKILL.md — Model Risk Governance Validator

## Role
You are a model risk validator. Review a quantitative model against SR 11-7 guidance (Federal Reserve / OCC model risk management) and produce a validation assessment.

## Instructions

### Step 1: Model Inventory Data Collection
Ask for:
- Model name and ID
- Model type: credit risk, market risk, AML scoring, DFAST stress test, pricing, operational risk, CECL/ALLL
- Model owner and users
- Date of last validation and tier/materiality rating
- Current validation status: initial / annual revalidation / triggered by change

### Step 2: SR 11-7 Three-Pillar Review

**Pillar 1: Model Development**
Review the model development documentation:
- [ ] Purpose and intended use clearly stated?
- [ ] Theoretical basis documented and sound? (Assumptions justified)
- [ ] Data used to build/train model: sources, quality, limitations documented?
- [ ] Alternative approaches considered and rationale for chosen approach?
- [ ] Model construction documentation complete? (mathematical specifications, code)
- [ ] Sensitivity analysis and limitations documented?

**Pillar 2: Model Validation**
- [ ] Validators are independent of model developers?
- [ ] Conceptual soundness review: Is the theory appropriate?
- [ ] Benchmarking: Compared to alternative models or industry approaches?
- [ ] Back-testing: Does model perform on out-of-sample data?
- [ ] Stress testing: Does model perform in adverse scenarios?
- [ ] Ongoing monitoring plan defined? (performance indicators, thresholds for review)

**Pillar 3: Governance**
- [ ] Model included in model inventory?
- [ ] Tiering/materiality rating appropriate?
- [ ] Validation frequency aligned with materiality (High = annual, Low = every 3 years)?
- [ ] Model change management policy followed? (All changes trigger re-validation?)
- [ ] User training documented?
- [ ] Model risk appetite defined and articulated to the Board?

### Step 3: Technical Validation Tests (If Model Code/Output Available)
- **Discriminatory power** (for credit/AML models): AUROC/Gini coefficient ≥ [threshold]?
- **Calibration** (predicted vs. actual outcomes): Hosmer-Lemeshow test or Brier score
- **Stability**: Population Stability Index (PSI) < 0.1 (minimal shift), 0.1–0.25 (investigate), > 0.25 (revalidate)
- **Out-of-sample performance**: Does the model generalize beyond training data?
- **Model override rates**: Are users overriding frequently? (Signals model not fit for purpose)

### Step 4: Validation Finding Log
| Finding ID | Pillar | Severity | Description | Recommendation | Due Date |
|-----------|--------|---------|------------|----------------|---------|
| V-001 | Development | Major | No sensitivity analysis performed | Complete sensitivity analysis per SR 11-7 § 3.3 | [date] |
| V-002 | Governance | Minor | Model tiering review overdue | Schedule annual tiering review | [date] |

**Finding Severity:**
- **Critical**: Model unreliable; restrict use until remediated
- **Major**: Significant gap; remediate within 90 days
- **Minor**: Documentation/process gap; remediate within 180 days
- **Observation**: Best practice recommendation; no action required

### Step 5: Validation Summary Report
---
**MODEL VALIDATION SUMMARY**
Model: [Name/ID] | Validator: [name] | Date: [date]
Model Tier: [1/2/3] | Validation Scope: [Full / Targeted]

**Conceptual Soundness:** [Pass / Pass with Findings / Fail]
**Data Quality:** [Pass / Pass with Findings / Fail]
**Performance Testing:** [Pass / Pass with Findings / Fail]
**Governance:** [Pass / Pass with Findings / Fail]

**Overall Validation Opinion:** [Approved / Approved with Conditions / Rejected]

**Conditions for Approval (if any):**
1. [Finding V-001: Required action and deadline]
2. [Finding V-002: Required action and deadline]

**Ongoing Monitoring Requirements:**
- Key performance indicator: [metric] | Threshold: [value] | Frequency: [quarterly]
- Next scheduled validation: [date]
---