Credit Risk Committee
Overview
Prepare committee-ready credit analysis for bank branches and credit-union segments using a shared credit office public API. The API provides branch data, loan portfolios, pending applications, sector exposures, credit policies, FDIC and NCUA benchmark data, and credit-union segment details. All analysis is driven by policy rules fetched from the API itself - never hard-code thresholds or formulas.
Quick Start
Always begin by loading the policy rules and confirming the target entity.
- Fetch
/api/policiesto load all credit rules, factor tables, and stress formulas. These are authoritative for every computation. - If the task targets a branch (
branch_id), fetch the branch, its metrics (most recent quarter at index 0), its loans, sector exposures, and applications as needed. - If the task targets a credit-union segment (
segment_id), fetch the segment and the NCUA benchmark table. - Fetch FDIC benchmarks when the task requires NPA or CRE delinquency comparison.
- Build the JSON answer from the bottom up: compute individual loan-level values first, then aggregate, then fill in composite summary objects.
Five Committee Workflows
1. Rating Migration Review
Re-derive risk ratings for loans meeting a current-rating threshold, then summarize migration, material downgrades, NPA benchmark variance, and the most severe problem credit. See Patterns 1-4 in workflow_patterns.md and credit_methodology.md.
Key data: branch loans (filtered by current_rating), branch metrics
(most recent quarter), FDIC benchmark. Start by re-deriving every loan's
final_rating using the dominant-factor rule, then build all aggregations
from those results.
2. Lending Committee Allocation
Screen pending applications against fatal-issue criteria, rank the survivors, allocate lending capacity in priority order, flag sector concentration breaches, and produce decline reason codes. See Pattern 6 in workflow_patterns.md.
Key data: branch (for lending_capacity_q1, sector_ceiling_pct),
applications, sector exposures. Compute committed_capacity_amount as the
bank-retained share (for SBA: approved * (1 - sba_guaranty_pct);
for participation: the retained portion).
3. Credit-Union Segment Posture
Compare a segment's state against NCUA national and peer-state benchmarks, determine a posture recommendation, and build operating controls and escalation triggers. See Pattern 7 in workflow_patterns.md.
Key data: credit-union segment, NCUA benchmark table. Compute peer median across the named peer states (exclude the target state and US row). Compare direction on all four metrics.
4. Watch-List Stress Packet
For loans with adverse ratings, assign CDFI risk classes from objective factors, run the +200bp DSCR stress, queue workout actions, and summarize by rating/payment buckets. See Pattern 5 in workflow_patterns.md.
Key data: branch loans filtered by current_rating threshold, policies for
CDFI factor tables and stress formula. The watch-list stress formula is
stressed_dscr = dscr / 1.18.
5. Competing CRE Decision
Compare two CRE applications with weighted CDFI scoring, CRE dual stress, branch CRE concentration analysis, and FDIC delinquency variance. Recommend the stronger credit with conditions and assign reason codes to the unselected one. See Pattern 8 in workflow_patterns.md.
Key data: the two applications, branch loans (for existing CRE exposure),
sector exposures, FDIC benchmark. The CRE dual stress formula is
stressed_dscr = dscr * 0.85 / 1.18.
API and Methodology References
- api_endpoints.md - Complete API reference with endpoint paths, field types, and data shapes. Load this first to understand what data is available.
- credit_methodology.md - Policy rules for risk-rating derivation, CDFI factor scoring, CRE weighted scoring, stress formulas, concentration limits, and benchmark comparison. Also covers watch-list action assignment and credit-union posture heuristics.
- workflow_patterns.md - Reusable computational patterns for each of the five committee workflows plus general numerical and ordering rules.
General Rules
Precision and Ordering
- USD exposure and currency: 2 decimal places
- Ratios: 4 decimal places unless the answer template specifies otherwise
- NCUA integer metrics (bps, pct): keep as integers
- Variance bps: 2 decimal places
- Weighted CDFI scores: 1 decimal place
- Sort lists as specified in each answer template's ordering rules
Answer Format
- Always output valid JSON matching the requested answer template shape
- Do not include narrative text outside the JSON
- Use only enum values and identifiers allowed by the template
- Include every required key; do not omit optional keys that have data
Data Fetching
- All API base URLs are supplied as
<TASK_ENV_BASE_URL>by the runner - No authentication is required
- Policy rules are always fetched fresh from
/api/policies- never hard-code thresholds or assume stale values - Use
/api/manifestfor endpoint discovery and record counts