Credit Office Committee JSON
Workflow
- Read the prompt, the matching
answer_template.json, and any endpoint names it mentions. - Fetch data only from the runner-provided
TASK_ENV_BASE_URLpublic API. - Identify the packet type, then follow the matching section in references/calculations.md.
- Preserve template key order, enum spelling, list ordering, and numeric precision.
- Return one JSON object only.
Packet Types
Branch Rating Migration
- Regrade the loans at or above the prompt threshold.
- Recompute final ratings from objective factors only, then summarize downgrades, NPA variance, and watch-list coverage.
Lending Allocation
- Score each pending application with the published weighted CDFI rubric.
- Approve only what fits capacity and concentration rules.
- Build the post-approval concentration view from branch sector exposures and latest total loans.
Credit-Union Segment Posture
- Use the segment record, NCUA benchmark row, and peer comparisons.
- Choose the posture, controls, escalation triggers, and interpretation directly from the data.
Watch-List Stress
- Treat the prompt’s adverse-rating floor as the watch-list population.
- Compute the +200bp DSCR stress, queue workout actions, and summarize severe buckets.
Competing CRE Decision
- Compare the two CRE requests with weighted score, stressed coverage, branch CRE concentration, and FDIC variance.
- Select one path and assign controlled reason codes to the unselected request.
Output Rules
- Match the schema exactly.
- Use only allowed enums and identifiers.
- Sort grouped IDs alphabetically unless the template says otherwise.
- Round only at the end.
Reference
- Read references/calculations.md before calculating.
- Use scripts/credit_office_toolkit.py for shared fetch and math helpers.