Wealth Advisory Planning
Use the staged API and the local task files to produce one JSON object that matches the supplied answer template.
Workflow
- Read the prompt, request memo, and answer template.
- Extract
client_idandtask_idfrom the local files or pass them explicitly. - Query only the allowed task-env endpoints:
/api/clients/{client_id}/api/source-documents/api/retirement-accounts/api/life-insurance/api/trust-candidates/api/policies/tax/api/rmd-factors
- Resolve conflicts with source precedence:
- prefer
SIGNED_PROFILEfor household facts and goals - use
ATTORNEY_MEMOwhen it is the highest-priority legal/planning note - use
CUSTODIAN_EXPORTfor retirement-account balances and RMD settings - fall back to
CRM_NOTEonly when newer sources do not supply the field
- prefer
- Use scripts/advisory_planner.py to assemble the output when possible.
- Emit JSON only. Do not add prose.
Model Rules
- For Roth conversion tasks, set annual conversion capacity from the filing-status bracket target minus annual non-IRA income, clamp at zero, and limit the annual amount so total conversions do not exceed the traditional balance.
- Project Roth and traditional balances year by year. Apply each year’s conversion before that year’s RMD, subtract the RMD before growth, and tax the distribution at the marginal rate.
- For ILIT tasks, use the signed beneficiary count for annual exclusion capacity, compare it to the annual premium, and derive notice dates from the planned contribution date.
- For GRAT/CRAT tasks, calculate remainder values from the expected growth rate, term, and payout or annuity rate. Use the tax policy estate-tax rate for GRAT tax reduction and the charitable-deduction rate for CRAT deduction.
- For estate-liquidity plans, include both the ILIT math and the trust-transfer comparison, sort
action_setalphabetically, and keep source-resolution fields aligned with the controlling records.
Output Discipline
- Match enum values exactly as the template defines them.
- Round money to cents and dates to ISO
YYYY-MM-DD. - Keep extra fields only when they are part of the observed template family.
- Prefer the newest higher-priority source when multiple records disagree.