Private Wealth Advisory Structured Planning

Private Wealth Advisory — Structured Planning SOP

You are given a client (e.g. CLT-1003), a request_memo.md with engagement context (sometimes a planning horizon year), and an answer_template.json that defines the EXACT output schema. The only data source is a remote HTTP API. Return one JSON object only, no prose, USD rounded to cents, ISO dates YYYY-MM-DD.

0. Standard operating procedure (do this every task)

1. Read prompt.txt, payloads/request_memo.md, and payloads/answer_template.json. The template's analysis_type enum tells you which of the 4 analyses to run; the required_top_level_keys and fields map tell you every key, type, and allowed enum value. Mirror the template exactly — produce every required key, use only listed enum values.
2. Pull all relevant API data for the client (see §1).
3. Resolve conflicting source documents (see §2) BEFORE computing anything.
4. Run the analysis-specific formulas (see §4–§7).
5. Set task_id to the task's stable id (e.g. train_003 / test_003; infer from the task folder name; if unknown use train_00X/test_00X matching the client index), and client_id to the exact client id string.
6. Emit JSON. Numbers are JSON numbers (not strings), rounded to 2 decimals. Sort any action_set list ALPHABETICALLY.

1. API usage (base URL from environment_access.md, usually

<remote-env-url>)

Use curl -s via Bash. If you run helper scripts use python (not python3).

Always fetch the live tax policy and rmd-factors — do not hardcode the numbers below; they are illustrative of the observed shape and may change per task. Observed constants: gift exclusion 2025=19000 / 2026=20000; estate exemption 2025=13,990,000 / 2026=13,610,000; estate_tax_rate=0.40; conversion bracket targets MFJ=394600 / SINGLE=197300 / HOH=263500; max_crat_term_years=20; charitable_deduction_rate=0.35.

2. Conflicting-source resolution (THE central trap)

Each client has 3 profile docs that DISAGREE. The stale CRM_NOTE is a distractor — it often inflates philanthropic_intent to "high" or lowers beneficiary_count, which would flip the recommendation if you trusted it. Resolve by source priority, not by reading whichever doc you saw first.

Priority (highest → lowest), matching the enum order in the templates:

SIGNED_PROFILE  >  ATTORNEY_MEMO  >  CUSTODIAN_EXPORT  >  CRM_NOTE  >  STALE_MARKETING_INTAKE

Tiebreak within equal priority: newer effective_date wins. (Observed dates: CRM_NOTE 2025-11-20, ATTORNEY_MEMO 2026-01-18, SIGNED_PROFILE 2026-02-06 — so recency and priority agree, but always apply PRIORITY first.)

Field-class controllers (what source_resolution.* must report):

• Profile / goal / beneficiary facts (annual_non_ira_income, marginal_tax_rate, beneficiary_count, philanthropic_intent, family_transfer_priority, age, filing_status, liquid_assets): SIGNED_PROFILE controls when present (it is the most complete doc). Report controlling_profile_source / controlling_goal_source / controlling_beneficiary_source = SIGNED_PROFILE.
• Account facts (traditional/roth balances, returns, rmd_start_age): the retirement-accounts export → controlling_account_source = CUSTODIAN_EXPORT.
• Policy facts (death_benefit, annual_premium, transfer flag): the life-insurance export. For controlling_policy_source choose the export source if the policy data comes from the custodian/insurer export, else SIGNED_PROFILE. Default to CUSTODIAN_EXPORT when the value originates from the life-insurance endpoint.
• Asset facts for trust comparison (estate_value, trust asset_value): estate_value appears in both ATTORNEY_MEMO and SIGNED_PROFILE (equal value); the trust asset_value comes from trust-candidates. For controlling_asset_source, prefer ATTORNEY_MEMO if its enum is offered and it supplies the estate/asset figure, otherwise SIGNED_PROFILE.

Rule of thumb: take each needed field from the highest-priority document that actually contains it. SIGNED_PROFILE carries nearly everything; fall back to ATTORNEY_MEMO then CRM_NOTE only for fields SIGNED_PROFILE omits.

3. Rounding, dates, enums

• Round every USD field to 2 decimals (cents). Keep year fields as integers.
• ISO dates YYYY-MM-DD.
• Use ONLY enum values listed in the template field descriptions. Never invent.
• action_set (estate plan) MUST be alphabetically sorted.

4. analysis_type = roth_conversion_rmd

Inputs: client (age, filing_status, planning_year), SIGNED_PROFILE (marginal_tax_rate, annual_non_ira_income), retirement export (traditional_balance, roth_balance, expected_return r, rmd_start_age, recommended_conversion_years), tax policy (conversion_bracket_targets), rmd-factors. Horizon year comes from the memo.

Definitions and formulas:

• first_rmd_year = planning_year + (rmd_start_age - age).
• years_until_rmd = first_rmd_year - planning_year.
• conversion_years = recommended_conversion_years.
• conversion_years_positive = min(conversion_years, max(0, years_until_rmd)) — the conversion years that actually fit BEFORE RMDs begin.
• first_conversion_year = planning_year.
• Bracket headroom: `headroom = conversion_bracket_targets[filing_status]
  • annual_non_ira_income`.
• annual_conversion_amount = min(headroom, traditional_balance / conversion_years) (limit conversions so taxable income stays under the bracket target; never exceed the balance pace).
• total_converted = annual_conversion_amount * conversion_years_positive (you can only convert in the years before RMD; this keeps the legacy projection self-consistent). If a task's numbers imply the full conversion_years were used, compute both and prefer the one consistent with the RMD simulation below.
• total_conversion_tax = total_converted * marginal_tax_rate.

RMD projection — simulate year by year from planning_year to horizon_year (inclusive). For each year grow the traditional balance by (1+r), then if current_age >= rmd_start_age withdraw RMD = balance / rmd_factor[age], tax it at marginal_tax_rate, and subtract it. Run TWO scenarios:

• Baseline (no conversion): accumulate baseline_rmd_tax_through_horizon.
• Conversion: in years first_conversion_year .. first_conversion_year + conversion_years_positive - 1, move annual_conversion_amount from traditional → roth (capped at remaining traditional) BEFORE growth; grow both accounts; take RMDs only on the (smaller) traditional balance. Accumulate conversion_rmd_tax_through_horizon.
• rmd_tax_savings_through_horizon = baseline_rmd_tax_through_horizon - conversion_rmd_tax_through_horizon (positive = conversion helps).

Legacy projection (use the conversion scenario's end balances at horizon):

• projected_roth_balance_horizon = roth balance at horizon.
• projected_traditional_balance_horizon = traditional balance at horizon.
• heir_tax_profile: ratio roth/(roth+trad) at horizon → MOSTLY_TAX_FREE if roth fraction ≳ 0.6, MOSTLY_TAXABLE if ≲ 0.3, else MIXED_TAXABLE_AND_TAX_FREE.

Recommendation:

• primary_action: NO_CONVERSION if headroom <= 0 (no room to convert in bracket); DEFER if the window is essentially gone (years_until_rmd <= 0, i.e. already at/over RMD age with negligible benefit); otherwise STAGED_ROTH_CONVERSION when rmd_tax_savings_through_horizon > 0.
• suitability: SUITABLE when staged conversion with clear savings and the conversion tax is comfortably covered by liquid assets; DEFER when primary_action is DEFER; else BORDERLINE.
• risk_flag: RMD_NEAR_TERM if the client is at/near RMD age (years_until_rmd <= 1); LIQUIDITY_CONSTRAINT if cumulative conversion tax approaches/exceeds liquid_assets; otherwise TAX_BRACKET_MANAGEMENT.

source_resolution: controlling_profile_source = SIGNED_PROFILE, controlling_account_source = CUSTODIAN_EXPORT.

5. analysis_type = ilit_crummey_implementation

Inputs: client (planning_year, estate_value), SIGNED_PROFILE (beneficiary_count — NOT the CRM count), life-insurance (death_benefit, annual_premium, planned_contribution_date, is_existing_policy_transfer), tax policy (annual_gift_exclusion, estate_tax_exemption, estate_tax_rate).

Gift plan:

• planning_year from client.
• annual_exclusion_per_beneficiary = annual_gift_exclusion[planning_year].
• beneficiary_count = SIGNED_PROFILE beneficiary_count.
• annual_exclusion_capacity = annual_exclusion_per_beneficiary * beneficiary_count.
• annual_premium from policy.
• premium_gap: the shortfall of exclusion vs premium. Report max(0, annual_premium - annual_exclusion_capacity) (a "gap" is non-negative; if exclusion fully covers the premium the gap is 0). Keep the raw signed value internally to drive the risk flag.

Administration (Crummey 30-day withdrawal window):

• contribution_date = policy planned_contribution_date.
• notice_due_date = the contribution date (Crummey notices issued at/by the contribution date).
• withdrawal_window_end = contribution_date + 30 days.
• earliest_premium_payment_date = withdrawal_window_end + 1 day (premium paid only after the withdrawal window closes).
• notices_required = beneficiary_count (one notice per Crummey beneficiary).
• dedicated_bank_account_required = true (ILIT best practice; a separate trust account preserves the gift/Crummey formalities).

Estate result:

• death_benefit from policy.
• estate_inclusion_risk (= recommendation.risk_flag), decided by two flags:
  • lookback = is_existing_policy_transfer (transferring an EXISTING policy triggers the IRC 2035 three-year lookback).
  • shortfall = annual_premium > annual_exclusion_capacity.
  • Combine: both → THREE_YEAR_LOOKBACK_AND_EXCLUSION_SHORTFALL; only transfer → THREE_YEAR_LOOKBACK; only shortfall → EXCLUSION_SHORTFALL; neither → LOW_IF_FORMALITIES_MET.
• projected_outside_estate_if_implemented = death_benefit if the policy is newly issued by the ILIT (no lookback); if it's an existing-policy transfer under the 3-year lookback, treat it as still includible → 0 (or note the lookback) until 3 years pass.
• tax_liquidity_support = the portion of the death benefit usable to cover estate tax = min(death_benefit, estate_tax_exposure) where estate_tax_exposure = max(0, estate_value - estate_tax_exemption[year]) * estate_tax_rate. (If the policy fully covers exposure, this equals exposure; otherwise the death_benefit.)

Recommendation:

• primary_action: if no shortfall and no lookback → FUND_WITH_CRUMMEY_NOTICES; if shortfall only → USE_LIFETIME_EXEMPTION_FOR_SHORTFALL; if lookback (existing transfer) → USE_NEW_POLICY_OR_ACCEPT_LOOKBACK; if both → DISCLOSE_LOOKBACK_AND_USE_EXEMPTION.
• suitability: SUITABLE_WITH_ADMINISTRATION when risk is LOW_IF_FORMALITIES_MET; BORDERLINE for a single issue; NOT_SUITABLE only if structurally broken.
• risk_flag = the estate_inclusion_risk computed above.

source_resolution: controlling_beneficiary_source = SIGNED_PROFILE, controlling_policy_source = CUSTODIAN_EXPORT (life-insurance export).

6. analysis_type = trust_comparison (GRAT vs CRAT)

Inputs: SIGNED_PROFILE goals (family_transfer_priority, philanthropic_intent), client/ATTORNEY estate_value & liquid_assets, trust-candidates (asset_value V, expected_growth_rate g, grat_term_years Tg, grat_annuity_rate ag, crat_term_years Tc, crat_payout_rate pc), tax policy.

Estate context:

• taxable_estate = max(0, estate_value - estate_tax_exemption[year]).
• estate_tax_exposure = taxable_estate * estate_tax_rate.
• liquidity_gap_before_planning = max(0, estate_tax_exposure - liquid_assets).

GRAT (annuity returns to grantor; remainder passes to heirs after term):

• annual annuity A = V * ag.
• FV_assets = V * (1+g)^Tg.
• FV_annuity = A * ((1+g)^Tg - 1) / g (ordinary-annuity future value).
• projected_remainder_to_heirs = FV_assets - FV_annuity.
• estimated_estate_tax_reduction = projected_remainder_to_heirs * estate_tax_rate (value removed from the taxable estate).
• mortality_inclusion_risk = TERM_SURVIVAL_REQUIRED (grantor must outlive the GRAT term or assets are pulled back into the estate).

CRAT (payout to income beneficiary for term; remainder to charity):

• payout P = V * pc; term Tc (cap at max_crat_term_years).
• FV_assets = V * (1+g)^Tc; FV_payout = P * ((1+g)^Tc - 1)/g.
• projected_charitable_remainder = FV_assets - FV_payout.
• estimated_income_tax_deduction = V * charitable_deduction_rate (charitable deduction approximated as asset value × the policy deduction rate).
• family_transfer_fit: CRAT sends the remainder to charity, so its fit for FAMILY transfer is LOW (use MODERATE/HIGH only if the data clearly shows family-directed remainder, which it normally does not for a CRAT).

Recommendation (driven by the CONTROLLING SIGNED_PROFILE goals, NOT the stale CRM philanthropic_intent):

• If philanthropic_intent == "high" → preferred_strategy = CRAT, rationale_code = PHILANTHROPIC_PRIORITY, alternate_role = SECONDARY_FAMILY_TRANSFER_TOOL.
• Otherwise (family_transfer_priority high and philanthropy not the top goal) → preferred_strategy = GRAT, rationale_code = CHILDREN_TRANSFER_PRIORITY, alternate_role = SECONDARY_CHARITABLE_TOOL.
• IMPORTANT TRAP: CRM_NOTE may say philanthropic_intent="high" while the controlling SIGNED_PROFILE says "moderate"/"low". Use SIGNED_PROFILE → this usually yields GRAT. Do not be misled by the CRM into choosing CRAT.

source_resolution: controlling_goal_source = SIGNED_PROFILE, controlling_asset_source = ATTORNEY_MEMO (or SIGNED_PROFILE if ATTORNEY_MEMO lacks the figure).

7. analysis_type = estate_liquidity_action_plan (integrated)

Combines §5 (ILIT) and §6 (trust) plus an action set. Inputs: client, SIGNED_PROFILE goals, life-insurance, trust-candidates, tax policy.

Estate context: same three formulas as §6 (taxable_estate, estate_tax_exposure, liquidity_gap_before_planning).

ilit block:

• annual_exclusion_capacity = annual_gift_exclusion[year] * beneficiary_count (SIGNED_PROFILE count).
• premium_gap = max(0, annual_premium - annual_exclusion_capacity).
• estate_inclusion_risk: same flag logic as §5 (lookback ∧/∨ shortfall).
• projected_outside_estate_if_implemented = death_benefit if no 3-year lookback, else 0.

trust_transfer block: run the §6 GRAT and CRAT math.

• preferred_strategy per §6 goal logic (usually GRAT).
• projected_remainder_to_heirs = GRAT remainder.
• estimated_estate_tax_reduction = GRAT remainder × estate_tax_rate.
• projected_charitable_remainder = CRAT remainder (report alongside).

recommendation:

• primary_action: if both an ILIT and a family-transfer GRAT are warranted → COMBINE_ILIT_AND_GRAT; if philanthropic priority dominates → CRAT_WITH_LIQUIDITY_REVIEW; if only the ILIT plus exemption work is needed → ILIT_WITH_EXEMPTION_REVIEW.
• sequencing: ILIT_FIRST_THEN_GRAT for the combined case; TRUST_DECISION_FIRST when the trust choice is the open question; ILIT_FIRST_THEN_ATTORNEY_REVIEW when legal drafting must follow funding.
• risk_flag: the ILIT estate_inclusion_risk flag.

action_set: list, alphabetically sorted, drawn ONLY from: ATTORNEY_DRAFT_REVIEW, CRAT_FOR_CHARITABLE_REMAINDER, GRAT_FOR_APPRECIATING_SHARES, ILIT_CRUMMEY_NOTICE_CYCLE, LIFETIME_EXEMPTION_ALLOCATION. Include the items that apply: ILIT funding → ILIT_CRUMMEY_NOTICE_CYCLE; appreciating-asset transfer via GRAT → GRAT_FOR_APPRECIATING_SHARES; charitable remainder considered → CRAT_FOR_CHARITABLE_REMAINDER; premium gap or estate above exemption → LIFETIME_EXEMPTION_ALLOCATION; any trust drafting → ATTORNEY_DRAFT_REVIEW. Always sort the final list alphabetically before output.

source_resolution: controlling_goal_source = SIGNED_PROFILE, controlling_policy_source = CUSTODIAN_EXPORT.

8. Common pitfalls / self-checks

• Do not trust CRM_NOTE. It is the stale distractor. SIGNED_PROFILE controls beneficiary_count, philanthropic_intent, family_transfer_priority, marginal_tax_rate, income. Trusting CRM flips GRAT↔CRAT and the bene count.
• premium_gap is a gap, not a signed difference — report max(0, …).
• conversion_years vs conversion_years_positive: only the positive (pre-RMD) years actually convert; keep total_converted and the legacy/RMD simulation consistent with conversion_years_positive.
• RMD timing: first RMD in the year the client reaches rmd_start_age (typically 73). Use the live rmd-factors map; for ages beyond the map clamp to the last available factor.
• Estate exposure uses the planning-year exemption, and rate is the policy estate_tax_rate (0.40). liquidity_gap is floored at 0.
• Three-year lookback is tied to is_existing_policy_transfer == true; a newly issued ILIT policy has NO lookback and the full death benefit sits outside the estate.
• Output discipline: single JSON object, every required key present, enums exact, USD to cents, dates ISO, action_set alphabetical, task_id/ client_id strings. No prose, no markdown, no trailing commentary.
• Always re-fetch tax policy and rmd-factors per task — never assume the constants above are current for the specific task instance.

9. Minimal worked sanity numbers (method illustration, NOT answers)

These show the SHAPE of correct math; recompute from live data each time.

• Roth (MFJ, income 185k, target 394.6k): headroom = 209,600; with 7 conversion years and a 2.8M IRA, annual_conversion = min(209,600, 400,000) = 209,600; total_conversion_tax = 7×209,600×0.32. RMD savings come from the two-scenario year-by-year simulation through the horizon (baseline minus conversion).
• ILIT (4 beneficiaries, 2026): capacity = 20,000×4 = 80,000; premium 78,000 → premium_gap = 0, risk = LOW_IF_FORMALITIES_MET; notice window = contribution_date+30d, earliest premium = +1 more day.
• Estate (estate 24.6M, 2026 exemption 13.61M): taxable = 10.99M, exposure = 4.396M, liquidity_gap = exposure − liquid_assets (floored at 0).
• GRAT (V=8M, g=8%, Tg=5, ag=4%): A=320k; FV_assets=8M×1.08^5; FV_annuity=320k×((1.08^5−1)/0.08); remainder = FV_assets − FV_annuity; estate_tax_reduction = remainder×0.40.