This server works best when prompts are explicit about the dataset, time basis, geography, and ranking criteria. This page covers prompt patterns and conventions. For copyable examples, see Usage Examples.

Prompting Principles

• Name the institution, state, or peer group clearly. Ambiguous prompts produce ambiguous results.
• State the date basis when using quarterly or annual FDIC data.
• Ask for ranking criteria explicitly when comparing institutions (e.g., “rank by ROA” not “find the best banks”).
• Prefer one analysis question per prompt unless you want the model to chain tools.
• Distinguish between quarterly and annual data — they have different cadences and date fields.

Date and Data Cadence Rules

Understanding the data cadence is essential for well-formed prompts.

Dataset	Cadence	Date Field	Example
Call Reports (financials)	Quarterly	REPDTE (YYYYMMDD)	20241231 for Q4 2024
Demographics	Quarterly	REPDTE (YYYYMMDD)	20241231 for Q4 2024
Summary data	Annual	YEAR	2024
Summary of Deposits (SOD)	Annual (June 30)	YEAR	2024 = as of June 30, 2024

Dollar amounts are in thousands unless otherwise noted. Publication lag is approximately 90 days after the reporting period.

Date rules for prompts

• You can describe dates naturally (“December 31, 2024” or “Q4 2024”). The model translates to REPDTE format.
• Financial and demographics prompts use quarterly dates. Specify quarter-end dates: 0331, 0630, 0930, 1231.
• Summary and SOD prompts use annual dates. SOD data is always as of June 30.
• Do not mix quarterly and annual data in a single prompt unless the prompt acknowledges the different date bases. For example: “Show quarterly financials for 2024 and also the June 2024 SOD deposits — note these reflect different reporting periods.”

Sample Prompts by Category

Institution discovery

Find active banks in California with over $10 billion in assets.

Tools used: fdic_search_institutions

What to expect: A filtered list of California institutions above the asset threshold, with name, location, charter class, and asset size.

Direct institution lookup

Get the FDIC institution record for CERT 3511.

Tools used: fdic_get_institution

Quarterly financials

Show quarterly financials for CERT 3511 during 2024, sorted newest first.

Tools used: fdic_search_financials

What to expect: Up to four quarterly Call Report records (Q1–Q4) with balance sheet, income, capital, and ratio data.

Annual summary data

Show annual summary data for CERT 628 from 2018 through 2024.

Tools used: fdic_search_summary

What to expect: Yearly aggregate snapshots — a different dataset from quarterly financials.

Branch-level deposits

Show all branch deposit totals for CERT 3511 from the 2024 Summary of Deposits report.

Tools used: fdic_search_sod

What to expect: Branch-by-branch deposit balances as of June 30, 2024.

Snapshot comparison across time

Compare active North Carolina banks between December 31, 2022 and December 31, 2024. Rank by deposit growth percentage, return the top 10.

Tools used: fdic_compare_bank_snapshots

What to expect: A ranked comparison showing growth and profitability changes across the two dates.

Peer group benchmarking

Build a peer group for CERT 29846 as of December 31, 2024 and rank it on ROA, efficiency ratio, and loan-to-deposit ratio.

Tools used: fdic_peer_group_analysis

What to expect: The subject ranked against auto-derived peers with percentile positions and peer medians.

Health assessment

Run a full health assessment for CERT 3511 with 8 quarters of trend history.

Tools used: fdic_analyze_bank_health

What to expect: A public_camels_proxy_v1 assessment with composite and component ratings, capital categorization, management overlay, trend analysis, and risk signals. This is an analytical proxy — not an official CAMELS rating.

Risk screening with follow-through

Scan all active banks in Wyoming for risk signals. For any bank with critical signals, also run a full health assessment.

Tools used: fdic_detect_risk_signals, fdic_analyze_bank_health

What to expect: A severity-ranked list of flagged institutions, with follow-up health assessments for the most concerning.

Multi-Tool Analysis Patterns

These prompts intentionally chain multiple tools for deeper analysis:

Compare active North Carolina banks between December 31, 2021 and June 30, 2025. Rank by deposit growth, top 10, and call out which also improved ROA.

Build a peer group for CERT 29846, then compare the bank to peer medians on ROA, efficiency, and loan-to-deposit.

Which bank failures since 2008 had the highest losses? For the top 3, show their pre-failure financials and risk signals.

For structured multi-tool workflows without manual orchestration, consider the Claude Code skills.

The Public Off-Site Proxy Model

The health and risk tools share a public_camels_proxy_v1 analytical engine:

• fdic_analyze_bank_health returns a full proxy assessment (overall band, component scores, risk signals, trend insights).
• fdic_compare_peer_health returns per-institution summary scores plus a full proxy for the highlighted subject.
• fdic_detect_risk_signals returns per-institution risk signals generated by the proxy engine.

These outputs are analytical proxies based on public FDIC data — not official CAMELS ratings, confidential supervisory conclusions, or investment advice. Management (M) is inferred from patterns, not examination findings. Sensitivity (S) uses proxy metrics.

Prompting Pitfalls

• “Find the best banks” — too vague. Say what “best” means (highest ROA, strongest capital, etc.).
• “Use latest branch data and latest financials” — can mix annual and quarterly sources unintentionally.
• “Compare all banks” — may be too broad. Add geography or peer filters.
• “What happened last quarter?” — specify the quarter-end date explicitly.

Recommended Follow-On Prompts

After an initial analysis, these follow-ups deepen insight:

• “Which of the top growers also improved profitability?”
• “Show the same peer group but sort by efficiency ratio instead.”
• “Run a health assessment for the lowest-ranked bank in that peer group.”
• “What does the management overlay say about the weakest bank?”
• “Compare the health scores of the top 5 and bottom 5 in that group.”