The /fdic-examiner-overlay command guides an experienced bank examiner or financial analyst through enriching a public_camels_proxy_v1 baseline with qualitative knowledge not available from public Call Report data.

This is a Claude Code skill, not an MCP tool. It requires Claude Code with the plugin installed. If you are using another MCP client, you can run fdic_analyze_bank_health to get the public-data baseline, but the structured overlay workflow is only available through this skill.

When to Use It

• You have examiner-grade qualitative knowledge (examination findings, classified asset details, management assessments) that you want to systematically layer onto a public-data baseline.
• You need a blended assessment with clear provenance separation between public-data findings and examiner inputs.
• You want bounded, auditable score adjustments rather than ad-hoc narrative overlay.

When Not to Use It

• You only need the public-data baseline: Use fdic_analyze_bank_health directly.
• You want a comprehensive institution report: Run Bank Deep Dive first, then use this skill to overlay examiner knowledge on the health assessment.
• You are screening a portfolio: Use Portfolio Surveillance for population-level screening.

Inputs

Input	Required	Description
Institution identity	Yes	Bank name or FDIC CERT number
Overlay domains	Interactive	Selected during the session from: management, asset_quality, earnings, liquidity_funding, sensitivity
Per-domain inputs	Interactive	Direction (upgrade/downgrade/confirm), confidence level, evidence basis, and domain-specific field responses

Capital is excluded from overlay in v1 because it is already the most rules-based component of the proxy.

Example Session

Step 1 — Invoke the skill:

/fdic-examiner-overlay

The skill asks which institution to assess, fetches the public-data baseline via fdic_analyze_bank_health, and presents the CAMELS-proxy scores.

Step 2 — Select domains to overlay:

management, asset_quality

Step 3 — Provide inputs for each domain. The skill asks structured questions: direction, confidence, evidence, and domain-specific fields (e.g., management tenure stability, board governance, internal controls).

Step 4 — Review proposed adjustments. The skill shows baseline scores, raw adjustments, confidence caps, and final adjusted scores. You confirm or revise.

Step 5 — Receive the blended output with seven sections and explicit provenance separation.

What Output to Expect

Section	Contents
1. Public Proxy Baseline	Institution identity, report date, overall band/score, component scores, management overlay, risk signals
2. Examiner Overlay Inputs	Per-domain: direction, confidence, evidence, source date, field responses
3. Overlay Adjustments	Per-domain: baseline score, raw adjustment, confidence cap, final adjustment, adjusted score
4. Blended Assessment	Adjusted component scores, overall score/band, change from baseline
5. Caveats	Data quality flags, mandatory provenance separation, domain-specific caveats
6. Exam Narrative	Structured paragraphs attributed to source layer
7. Structured Worksheet	Machine-readable JSON appendix for audit trail

Provenance Separation

Every paragraph in the exam narrative is attributed to its source:

• “Public data indicate…“ — findings from the proxy baseline
• “Examiner overlay:” — analyst-provided qualitative inputs
• “Blended interpretation:” — combined assessment drawing on both sources

The three layers are never merged into undifferentiated prose.

How Adjustments Work

Evidence Magnitude

Each field response maps to an evidence magnitude (0.00 = no basis, 0.25 = moderate, 0.50 = strong) that quantifies the strength of the adjustment basis.

Direction and Confidence

• Upgrade: Proxy underrates the domain; examiner has evidence of strength.
• Downgrade: Proxy overrates the domain; examiner has evidence of weakness.
• Confirm: Examiner agrees with the proxy. No score change, but evidence is recorded for the audit trail.

Confidence caps the maximum adjustment: low (±0.50), medium (±0.75), high (±1.00).

Severe Overrides

Certain field responses produce a minimum adjustment floor that cannot be diluted (downgrade only): enforcement actions, deficient internal controls, absent contingency funding, IRR limits exceeded, or inadequate allowance. Each carries a 0.50 minimum floor.

Management Path

Management does not produce a component score adjustment. Instead, it modifies the management overlay state (normal / watch / elevated_concern). At elevated_concern, the final overall band is demoted one level.

Recomputation

Adjusted component scores are weighted (Capital 0.30, Asset Quality 0.25, Earnings 0.20, Liquidity 0.15, Sensitivity 0.10) and mapped to bands using the same thresholds as the proxy.

Overlay Domains

Domain	What the Proxy Sees	What the Examiner Adds
Management	Algorithmic overlay from component ratings and trends	Tenure stability, governance, controls, risk management, succession, supervisory history
Asset Quality	NCL ratio, charge-offs, reserve coverage, noncurrent assets	Concentration risk, underwriting quality, classified trends, allowance adequacy
Earnings	ROA, ROE, NIM, efficiency ratio	Sustainability, diversification, expense management
Liquidity/Funding	Loan-to-deposit, core deposits, brokered deposits, cash ratio	Contingency plans, borrowing capacity, depositor concentration, stability
Sensitivity	Rate risk proxy from long-term assets and volatile liabilities	IRR model results (EVE/NII), hedging effectiveness, rate exposure

Key Caveats

• Claude Code only: The interactive overlay workflow is not available in other MCP clients.
• Examiner knowledge required: The skill collects specific, structured inputs. Vague evidence will be re-prompted.
• Bounded adjustments: No single domain can move a score by more than ±1.0. Low-confidence inputs are further capped.
• Confidential inputs: Fields like classified asset trends and supervisory history may contain confidential supervisory information. The caveats section marks these. Handle the blended output according to your institution’s information-sharing policies.
• Capital excluded: Capital overlay is not available in v1.
• Proxy baseline: The starting point is a public-data analytical proxy, not an official CAMELS rating.

Planned Overlay Skills

Skill	Status	Domain
fdic-examiner-overlay	v1 available	Management, asset quality, earnings, liquidity/funding, sensitivity
fdic-liquidity-funding-depth	Planned	Contingency funding, borrowing capacity, depositor concentration
fdic-interest-rate-sensitivity-overlay	Planned	EVE/NII results, hedging, repricing exposure
fdic-concentration-risk-context	Planned	CRE, sector, single-borrower, geographic concentration
fdic-exam-narrative-generator	Planned	Examination-quality narrative from proxy and overlay outputs

Promotion Path

The examiner overlay is designed to stabilize through analyst use before potential promotion to a programmatic MCP tool. Promotion criteria: stable field enums, consistent scoring, stable worksheet JSON shape, and validation through real examination workflows.

Under the Hood

The skill orchestrates one MCP tool:

Tool	Purpose
fdic_analyze_bank_health	Fetch the public_camels_proxy_v1 baseline for the target institution

All other computation (evidence magnitudes, adjustment caps, severe overrides, recomputation) happens within the skill’s conversational logic, not through additional MCP tool calls.