Credit Office Committee JSON
Use this skill when a task asks for a committee-ready JSON answer based on the shared credit office public API and an input/payloads/answer_template.json file.
Core Workflow
- Read the task prompt and the answer template before fetching data.
- Identify the target object:
branch_id,segment_id, namedapplication_ids, review/as-of date, benchmark period, adverse-rating threshold, and required JSON keys. - Use the task environment base URL from the runner or prompt. Fetch only public read endpoints:
/api/manifest/api/policies/api/benchmarks/fdic/q4-2024/api/benchmarks/ncua/q1-2025/api/branches/{branch_id}/api/branches/{branch_id}/metrics/api/branches/{branch_id}/loans/api/branches/{branch_id}/sector-exposures/api/branches/{branch_id}/applications/api/credit-union-segments/{segment_id}
- Treat
/api/policiesas authoritative for thresholds, weights, stress formulas, versions, and allowed mitigations. Do not hard-code benchmark values; fetch them. - Use the metric row matching the prompt date/quarter. For quarter-end dates like March 31, use
2025Q1; otherwise use the most recent row unless the prompt specifies another quarter. - Build exactly the template shape. Preserve required keys, enum spellings, requested ordering, precision, and JSON-only output.
Precision And Ordering
- Currency and exposure fields: round to 2 decimals.
- Ratios: round to the template precision, usually 4 decimals.
- Basis points:
(ratio_variance * 10000), rounded to 2 decimals when requested. - Lists sorted by identifier should use lexical ascending order.
- For grouped exposure totals, sum unrounded source values first, then round the emitted total.
- If a template says a field is a set, still emit a stable list ordered as the template choices or alphabetically when no template order is obvious.
Risk Rating Regrade
For branch rating migration reviews:
- Fetch branch details, metrics, loans, policies, and FDIC benchmark data.
- Select loans in the prompt population, usually
current_rating >= target_current_rating_min. - Re-derive each selected loan's final rating using the policy risk-rating rules:
- DSCR thresholds map to ratings from best to worst.
- LTV/collateral thresholds map to ratings from best to worst.
- Payment status applies policy delinquency minimums.
- Final rating is the worst numeric rating from all available DSCR, LTV/collateral, and delinquency factors. Ignore missing factors rather than assigning a penalty unless policy says otherwise.
- Aggregate final-rating exposure totals by final rating.
- For requested migrations from a starting rating, filter original
current_rating, group by final rating, and include sorted loan IDs. - Material downgrades are
final_rating - current_rating >= policies.risk_rating.material_downgrade_notches; sort by loan ID. - Watch-list follow-up coverage normally includes final ratings requiring action:
- rating 6:
watchlist - rating 7:
special_assets - rating 8 or clear projected-loss/nonaccrual case:
partial_chargeoff_review - rating 5 only when the prompt asks for monitor coverage:
monitor
- rating 6:
- Choose the top problem credit by most severe final rating, then most severe payment status, then exposure. Include borrower, exposure, current/final rating, payment status, and recommended action.
- NPA benchmark variance:
branch_npa_exposure = metrics.nonperforming_loansbranch_total_loans = metrics.total_loans_outstandingbranch_npa_ratio = branch_npa_exposure / branch_total_loansvariance_ratio = branch_ratio - fdic_metric_ratiovariance_bps = variance_ratio * 10000
Pending Application Allocation
For branch pending-application packages:
- Fetch branch details, latest metrics, sector exposures, applications, and policies.
- Evaluate each application for hard risk reasons using controlled template reason codes:
weak_dscr: DSCR below policy floor or stressed DSCR breaches the required threshold.high_ltv: LTV exceeds the relevant policy/collateral tolerance.low_fico: FICO is materially below policy consumer floor.recent_bankruptcy: recent bankruptcy is reported.startup_risk: operating history is short; SBA guaranty and monitoring can mitigate.documentation_gap:documentation_completeis false.sector_breach: approval worsens a sector or CRE concentration above its limit without mitigation.capacity_limit: bank-retained exposure cannot fit branch capacity or a capped sector.fdic_adverse_varianceorncua_peer_weakness: external benchmark weakness is material and the template permits the code.
- Use decisions conservatively:
approvefor applications passing credit, capacity, and concentration checks without special mitigation.conditional_approvewhen approval needs SBA guaranty, monitoring, reduced amount, board exception, or other template condition.participation_requiredwhen the full gross request can be approved only by limiting bank-retained exposure.deferwhen the credit is not selected in a competing allocation but is not an outright decline.declinefor hard policy failures or unmitigated capacity/concentration failures.
- Track both gross and retained exposure:
approved_amountis normally the gross request approved.bank_capacity_usedis the retained exposure after participation, guaranty, or reduction.- With an SBA guaranty, retained exposure is usually
requested_amount * (1 - sba_guaranty_pct). - With participation, cap retained exposure so pro-forma retained sector exposure stays at or below its limit when possible.
- Allocation totals:
gross_approved_amount = sum(approved_amount for approved/conditional/participation_required)committed_capacity_amount = sum(bank_capacity_used for approved/conditional/participation_required)remaining_capacity = branch.lending_capacity_q1 - committed_capacity_amount
- Post-approval concentration views generally use gross approved exposure:
- numerator: current sector exposure plus approved gross amounts in that sector
- denominator: latest total loans outstanding plus total gross approved amount
- compare to the sector-specific
limit_pctfrom sector exposures, falling back to branch default.
- Emit concentration flags for applications that breach, approach, or require handling for a sector limit. Sort as the template requires.
- Priority rankings include only approved, conditional, and participation-required applications. Rank stronger credits and committee priorities first; when unclear, prefer stronger DSCR, lower LTV, stronger guaranty/relationship, and mitigated concentration use.
CRE Competing Decisions
For tasks comparing CRE applications:
- Fetch the named applications, branch details, latest metrics, existing loans, sector exposures, policies, and FDIC benchmark data.
- Existing CRE exposure is the sum of existing loans with
loan_type == "CRE"; CRE concentration is existing CRE exposure divided by latest total loans. - Compute CRE dual-stress using the policy expression, typically
stressed_dscr = dscr * 0.85 / (1 + 0.18), and compare to the policy coverage threshold. - Compute weighted CDFI score using
policies.cre_weighted_score.weights. Lower is better.- Capacity: DSCR and stressed repayment coverage.
- Capital: debt-to-asset, deriving
total_debt / total_assetswhen needed. - Character: guarantor strength, prior delinquencies, relationship depth, bankruptcy, and documentation.
- Collateral/exposure: LTV and collateral support.
- Conditions: sector/CRE concentration, branch capacity, benchmark underperformance, and external conditions.
- Map weighted score to the policy classes: approve-quality up to the lower max, conditional up to the next max, weak above that.
- Add reason codes from objective failures: stressed DSCR breach, high LTV, sector/CRE breach, FDIC adverse variance, documentation gaps, and other template-approved reasons.
- Select the stronger application by lower score, stress pass, better collateral, and feasible concentration handling. If the selected credit still worsens an over-limit CRE book, use a conditional or participation path and include committee/retained-exposure conditions.
- CRE concentration for the selected path:
selected_post_approval_cre_concentration = (existing_cre_exposure + selected_gross_amount) / (latest_total_loans + selected_gross_amount)selected_policy_variance_bps = (selected_post_approval_cre_concentration - branch.cre_policy_limit_pct) * 10000
- FDIC delinquency variance usually compares branch
delinquency_30_plus_pcttototal_real_estate_30_89_pctfor CRE delinquency tasks.
Watch-List Stress And Workout Queue
For adversely rated loan watch-list packets:
- Select loans with
current_rating >= adverse_rating_minfrom the prompt. adverse_balanceis the selected loans' outstanding balance total.- Compute CDFI factor score from available objective factors using policy tables:
- debt-to-asset score
- LTV score
- liquidity-months score
- FICO score when present Missing factors contribute no points unless the prompt or policy says otherwise.
- Map factor score to CDFI class using policy class ranges. Override to
Projected Lossfor underwater collateral with nonaccrual or similarly clear loss facts when the template asks for projected-loss identification. - Watch-list DSCR stress uses the policy formula, typically
stressed_dscr = dscr / (1 + 0.18), with shock label from policy. Include only loans with DSCR available. breaches_thresholdis true when stressed DSCR is below the policy threshold;breach_loan_idsare sorted ascending.- Workout queue includes the selected adverse population sorted by descending exposure, then loan ID. Recommended actions:
- current/final rating 6:
watchlist - current/final rating 7 or 90+ days past due:
special_assets - rating 8, nonaccrual, projected loss, or underwater collateral with loss indication:
partial_chargeoff_review - use
workoutorlegal_referralonly when the prompt/data clearly points there and the template permits it.
- current/final rating 6:
- Severe bucket counts group selected loans by current rating and payment status; sort by current rating, then the template's payment-status ordering or lexical order if no order is provided.
- Monitoring cadence is
monthlywhen any selected loan is rating 7+, nonaccrual/projected loss, or breaches stress;quarterlyfor moderate watch-list populations;semiannualonly for stable low-severity populations.
Credit-Union Segment Posture
For segment posture pages:
- Fetch manifest, policies, NCUA benchmark data, and the target credit-union segment.
- Select the segment state row, the
USrow, and the segment's peer states from the NCUA benchmark. - For peer medians, sort the peer values and use the middle value for three peers; for even counts use the arithmetic median.
- Direction fields compare the segment state value to the comparison value:
higherif state value is greaterlowerif lessequalif equal
- Interpret external risk with metric polarity:
- Higher delinquency and loan-to-share are weaker.
- Lower ROAA and positive-net-income percentage are weaker.
- Weaker on most metrics versus both national and peers means
weaker_than_national_and_peers; mixed signs meanmixed_vs_national_and_peers; stronger signs meanstronger_than_national_and_peers.
- Capacity status:
- positive current quarterly capacity:
capacity_available - limited or nearly consumed capacity:
capacity_constrained - no remaining capacity:
no_capacity
- positive current quarterly capacity:
- Posture:
- Capacity available with weaker or mixed external risk usually supports
continue_with_tighter_conditions. - No capacity or severe external weakness supports
temporarily_pause. - Capacity available with strong external metrics and no control issues supports
continue_approving.
- Capacity available with weaker or mixed external risk usually supports
- Required checklist gates come from the segment
minimum_checklist, constrained to template choices. - Added operating controls should be tied to objective facts:
- insurance issue:
pre_close_insurance_binder_verification - lien/UCC/title issue:
lien_perfection_prior_to_funding - external benchmark weakness:
quarterly_state_benchmark_monitoring - elevated or near-threshold segment delinquency:
monthly_segment_delinquency_watch - staffing/complexity or tighter posture:
senior_underwriter_second_review - capacity overrun/exception:
committee_exception_for_capacity_overrun
- insurance issue:
- Escalation triggers use template IDs/owners. Include only fact-supported triggers, commonly recent segment delinquency near or above threshold, missing insurance/lien exceptions, capacity exceptions, and widening state delinquency gaps.
Final JSON Checks
Before finalizing:
- Re-open the answer template and verify every required key is present.
- Check enum values against the template exactly.
- Confirm branch/segment/application IDs come from the prompt/API, not memory.
- Recalculate every subtotal from source rows.
- Ensure arrays are in the required order.
- Return only valid JSON, with no markdown or explanatory text.