The L3 Ch1 design package is in place. The audit identified Q4 Roth conversion season as a top-three 90-day intervention. The handoff diagram specified propose-mode with three-tier verification and signoff under Reg BI Care Obligation. The verification specification defined the source-system, regulatory, and client-fit checkpoints with named systems, named statutes (IRC §408(d)(2) read with §72(e)(8) Form 8606 for pro-rata — not §408(d)(6) which governs divorce), and named household-state inputs. This lesson takes that design and runs the first L3 Ch2 workflow at scale: the 50-household Roth conversion screen. Pull every client over age 55 from Wealthbox, cross-reference RightCapital marginal tax projections and Holistiplan-extracted prior-year AGI, identify the households with conversion windows (pre-RMD, pre-Social Security, low-income years, post-business-sale), and rank-order by leverage — dollars of long-term tax saved per dollar of current-year tax paid. The deliverable is a prioritized screen output that becomes the input to the L3 Ch2 L2 per-client sizing memo and the L3 Ch2 L3 Reg BI recommendation, execution, and archiving workflow.

Why the 50-Household Screen Replaces the Dozen Most-Obvious

Pre-AI, the 200-household practice runs Roth conversion conversations with the dozen most-obvious candidates each Q4 — the clients who explicitly asked the year before, the clients whose 1040 the senior advisor remembers reading, the clients whose spouse mentioned Medicare planning at the last review. The L3 Ch1 L1 audit's typical finding is that the practice has another 30+ households where the leverage is comparable but the household wasn't surfaced by ad-hoc memory. The 50-household screen catches the missed cohort by running the leverage calculation systematically across the entire eligible population.

The eligibility filter is age 55+ as a practical floor (the conversion calculus rarely produces meaningful leverage below this age because the retirement window and the RMD-pressure timing aren't yet aligned), plus the conditional inclusions: clients with significant post-business-sale low-income windows, clients with current tax-bracket compression that opens an unusual conversion year, clients on the cusp of Social Security claiming who need to coordinate, and clients with inherited IRAs (10-year rule) where conversion math intersects with the inherited withdrawal schedule. The practical 200-household practice typically surfaces 45-60 eligible households when the filter runs systematically.

The Data Pull — Source-System Checkpoint at Scale

The screen's source-system checkpoint runs at population scale. For each eligible household, the AI workflow pulls: Wealthbox household record (age, spouse age if applicable, accounts inventory, beneficiary status, IPS edition reference); Holistiplan-extracted prior-year 1040 (AGI from Line 11, taxable income from Line 15, federal tax from Line 24, capital gains breakdown, itemized deduction state); RightCapital plan (current-year and next-three-year marginal tax projection, asset allocation, Monte Carlo position, Social Security claiming assumption, withdrawal-order strategy); aggregated pre-tax IRA basis across all custodians (Schwab, Fidelity, Pershing, BNY Mellon, Edward Jones, TD legacy/Schwab Advisor) for the §408(d)(2)+§72(e)(8) Form 8606 calculation; current Roth IRA balances and prior Form 8606 filings for basis-tracking; Social Security PIA from SSA portal or RightCapital plan assumption; client domicile from Wealthbox for state-tax calculation; current CMS IRMAA threshold table.

The source-system checkpoint passes when every household's data elements have named source, named tool, timestamp, and field-match. The discipline catches the data-quality issues that would otherwise contaminate the screen: stale Holistiplan extractions where the client filed an amendment; aggregated-basis errors where a custodian feed lags a recent rollover-in or contribution; Wealthbox age fields that haven't been updated for newly-65 spouses; Roth balance feeds missing a client's external Roth at a custodian the firm doesn't have a feed from.

The Conversion Window Typology

The AI workflow then classifies each eligible household by conversion window type. The five canonical windows the L3 Ch2 L1 screen surfaces.

Pre-RMD Window

Client is age 55-72 (under the SECURE 2.0 age-73 RMD trigger; rising to 75 in 2033). Pre-tax IRA balances are still growing tax-deferred; once RMDs begin, the mandated distributions push income into a brackets the client may want to avoid. Conversions in the pre-RMD window shrink the future-RMD base, reducing the bracket compression at RMD-age and beyond. The leverage calculation projects RMDs at age 73-75-80-85 under no-conversion vs partial-conversion vs aggressive-conversion scenarios and computes the lifetime tax-bracket impact. The pre-RMD window is the most common L3 Ch2 conversion candidate.

Pre-Social Security Window

Client is retired or semi-retired between, say, age 62 (early-claim eligibility) and 70 (delayed-claim cap). If the client delays SS to 70 to maximize benefit and survivor protection, the years between retirement and claim are often unusually low-income years where bracket-fill conversions cost less than they ever will again. The leverage calculation models the delay-SS-and-convert strategy vs claim-early-and-skip-conversion strategy across the household lifetime. The Hendersons example (married couple 64 and 62, the program's canonical example) falls squarely into this window for the higher-earner.

Low-Income Year Window

Client has an unusually low-income year — sabbatical, job transition, sale of underperforming business, recent retirement, recovery year from a divorce or medical event. The single-year tax bracket is materially lower than historical or projected average. The leverage calculation captures the one-time arbitrage and recommends conversion sizing to fill the gap to the next bracket without crossing into a punitive tier.

Post-Business-Sale Window

Client just sold a business or took a meaningful liquidity event. Year-of-sale income may be elevated (capital gain harvest) or compressed (deferred consideration smoothing out); post-sale year(s) often have artificially low income because business income ceased. The leverage calculation coordinates with the L3 Ch6 L4 business-sale planning workflow (QSBS §1202 dual regime, installment-sale §453, CRT funding pre-sale, opportunity zone §1400Z-2) and identifies the post-sale conversion-window timing.

Inherited-IRA Coordination Window

Client is a non-eligible designated beneficiary under SECURE 2.0 with an inherited IRA subject to the 10-year rule. Year-by-year withdrawals over the 10-year window need to smooth tax brackets; conversion of the client's own pre-tax balances can be coordinated to avoid double-stacking high-income years. The leverage calculation runs the 10-year window optimization (L3 Ch3 L2 next chapter) jointly with the client's own conversion schedule.

The Leverage Calculation — Long-Term Tax Saved Per Current-Year Tax Paid

The screen's central computational engine is the leverage ratio: dollars of long-term tax saved per dollar of current-year tax paid. The AI workflow computes leverage per eligible household by modeling no-conversion vs partial-conversion vs aggressive-conversion scenarios across the household's projected lifetime, summing the lifetime tax differential, and dividing by the current-year tax cost of the proposed conversion.

The components of the calculation: current-year conversion tax cost (federal bracket fill from current AGI to top of target bracket — typically 22% or 24% — at the client's marginal rate; NIIT 3.8% on any investment income above $200K single / $250K MFJ thresholds; state tax at client domicile rate; IRMAA two-year-lookback impact on Medicare premiums in year+2; any pro-rata-rule-induced tax-cost from aggregated pre-tax IRA basis per §408(d)(2)+§72(e)(8)). Long-term tax savings (reduced future RMDs out to age 95+ using the Uniform Lifetime Table or Single Life Table; reduced surviving-spouse single-filer bracket compression after first-death; reduced bequeath-IRA 10-year-rule beneficiary tax burden; reduced lifetime IRMAA-tier exposure; reduced lifetime state-tax exposure given likely domicile changes for retirement). Discount rate to present value (typically a real-rate assumption like 2-3%, sometimes the household's IPS-stated planning discount).

The leverage ratio output is a single decimal per household — e.g., the Hendersons might show leverage of 2.4 (meaning $2.40 of long-term tax saved per $1.00 of current-year tax paid on a $74K bracket-fill conversion). A client with leverage 4.0+ is a high-priority candidate; 2.0-4.0 is solid mid-priority; 1.0-2.0 is borderline; below 1.0 is generally not worth pursuing this year.

Leverage Calculation Edge Cases

Three edge cases the workflow handles explicitly. (a) Negative leverage: a household whose pro-rata-rule impact, combined with state-tax cost and IRMAA cliff exposure, makes any conversion negatively-leveraged this year. Output flags the household with the explicit reason and defers to a future year when conditions change. (b) Cliff-leverage: a household where small conversion is high-leverage but larger conversion crosses a tier (federal bracket, IRMAA, state tax) and goes sharply negative. Output identifies the optimal conversion size at the cliff edge, not the bracket-fill maximum. (c) Coordinated leverage: a household where the conversion competes with a concurrent planning workflow (capital gain harvest, NUA election, deferred-comp distribution under §409A, charitable bunching, QCD usage under §408(d)(8)). Output coordinates the joint optimization or flags the dependency for senior-advisor judgment.

The Prioritized Screen Output

The screen output is a ranked table of the 45-60 eligible households with: household_id, ages, current AGI, conversion window type, recommended conversion size (preliminary), current-year tax cost estimate, long-term tax savings estimate, leverage ratio, pro-rata-rule impact flag, IRMAA cliff flag, coordination dependencies, IPS-staleness flag, prior-year conversion history, source-system check pass-state, regulatory check pass-state, client-fit check pass-state, recommended-next-action (advance to L3 Ch2 L2 per-client sizing memo / defer to next year / coordinate with named workflow first / refer to senior advisor for judgment).

The table is structured per L2 Ch8 L2 (JSON object array, conforming to the firm's screen-output schema v2.4) so the L3 Ch2 L2 per-client sizing memo workflow ingests cleanly. The table is tagged-Markdown-routed to the Smarsh archive with metadata (document_type=roth_screen_q4_2026, batch_id, advisor IDs, retention policy). The senior advisor reviews the top of the ranked list (typically top 15-20 households by leverage), applies judgment on the recommended-next-action column, and authorizes the L3 Ch2 L2 sizing-memo workflow for the approved cohort. The bottom of the list (low or negative leverage) gets documented "deferred" with rationale and surfaces in the next quarterly screen run.

The Handoff Diagram and Checkpoint Chain in Action

The L3 Ch1 L2 handoff diagram for the screen workflow specifies: trigger = Q4 batch run (October 1 typically); inputs = Wealthbox household population, Holistiplan extractions, RightCapital plans, aggregated pre-tax IRA basis pulls, SSA portal / RightCapital PIA, current CMS IRMAA tables, current-year tax law; AI mode = propose (the screen output is a per-household proposed-next-action, not yet a recommendation); output = structured JSON array per schema v2.4; tool stack = Wealthbox + Holistiplan + RightCapital + custom-prompt GPT-4-class with firm system prompt v3.2 + screen-output schema validator; verification = three-tier per L3 Ch1 L3 verification specification (source-system per household, regulatory at workflow level, client-fit per household); signoff = senior advisor reviews the prioritized table, authorizes the cohort for L3 Ch2 L2 advancement, documents deferrals; downstream = L3 Ch2 L2 sizing-memo workflow for approved cohort, deferred-list back into Wealthbox household record for next quarterly screen; retention = Smarsh archive package with full inputs snapshot, AI output, verification logs, signoff narrative, advancement-cohort list, deferral list with rationale.

The checkpoint chain operates at scale. Source-system checkpoint runs per household and produces 45-60 discrete log entries (one per eligible household), each capturing pass/fail on the data-element list. Regulatory checkpoint runs at workflow level (verifying the regulatory framing applied uniformly — §408(d)(2)+§72(e)(8) for pro-rata, IRMAA 2024 lookback for 2026, current SECURE 2.0 ages) and produces one batch-level log entry. Client-fit checkpoint runs per household with the senior advisor's judgment narrative — the IPS-fit assessment, the prior-preference reconciliation, the coordinated-workflow check — and produces 45-60 discrete log entries. The complete retained checkpoint chain for the Q4 batch is roughly 100-130 log entries, all tagged for queryable surfacing.

The 50-Household Screen as Population-Level Discipline

The screen's operational power is that it converts an ad-hoc, memory-driven, dozen-household conversation into a systematic, evidence-driven, full-population workflow. The recovered hours measured against the L3 Ch1 L1 audit baseline are typically 80-130 hours per quarter for a 200-household practice (the pre-AI baseline was 4-6 hours of senior advisor research per household conversation times the dozen households the practice ran). The discovered cohort — the 30+ households the prior process missed — is where the L3 Ch2 deliverable becomes both client-value-generating and practice-revenue-defending.

The Marketing Rule 206(4)-1 implication for the screen output itself is contained: the screen is internal-only at this stage; no client-facing communication is produced. The client-facing artifacts emerge from L3 Ch2 L2 (per-client sizing memo) and L3 Ch2 L3 (Reg BI-compliant recommendation memo with the four-alternatives documentation). The screen stays inside the firm as the operational artifact that drives prioritization. The Reg BI implication is anticipatory — the screen's source-system and regulatory checkpoint logs become part of the eventual Reg BI evidence file for each household that advances to recommendation.

The CCO supervisory implication under FINRA Rule 3110 is review of the screen's verification log archive at quarterly sampling — the rate is typically lower than for the per-household L3 Ch2 L3 recommendations (because the screen is propose-mode at population scale, not yet propose-mode at recommendation scale), but the CCO checks that the screen ran on the right population, that the data sources were current, that the regulatory framing was correct, and that the deferred cohort's rationale was documented.

The Hand-Off to L3 Ch2 L2 Per-Client Sizing Memo

The screen output's approved cohort flows to L3 Ch2 L2's per-client sizing memo workflow. That lesson takes a single household at a time, runs the bracket-fill / NIIT / IRMAA two-year lookback (avoiding next-tier cliff) / state tax / pro-rata-rule check / five-year clock implications math, and produces the propose-mode recommendation memo for senior-advisor signoff. The Hendersons walk-through from L3 Ch1 L2 (recommend $74K conversion to preserve Tier 2 IRMAA position, coordinate with SS delay to 70, flag trust funding for L2 Ch5 L2 IPS update) is the L3 Ch2 L2 work product. The L3 Ch2 L3 lesson then takes the propose-mode memo and produces the Reg BI-compliant recommendation memo with execution (trade ticket, custodian conversion form, 1099-R expectations, CRM activity log, Smarsh archive, tax-prep handoff to client's CPA).

Every L3 Ch2 lesson is built on the L3 Ch1 design package — audit-driven prioritization, handoff-diagram-defined structure, verification-specification-driven checkpoints. The Q4 Roth conversion season's recovered hours, client-value-generating decisions, and Reg BI-defensible documentation come from the design-package discipline applied consistently across the three lessons.

Key Takeaways

• The 50-household screen replaces the dozen-most-obvious ad-hoc conversation. The L3 Ch1 L1 audit's typical finding is that the practice has another 30+ households with comparable leverage that ad-hoc memory misses; the screen catches them by running the leverage calculation systematically across the eligible population.
• Eligibility filter is age 55+ plus conditional inclusions (post-business-sale low-income windows, current-year tax-bracket compression, near-Social-Security-claiming, inherited-IRA 10-year-rule coordination). Typical 200-household practice surfaces 45-60 eligible households.
• Source-system checkpoint at population scale pulls Wealthbox household records, Holistiplan-extracted 1040s (AGI Line 11, taxable income Line 15, federal tax Line 24), RightCapital plans, aggregated pre-tax IRA basis across all custodians for the §408(d)(2)+§72(e)(8) Form 8606 calculation (not §408(d)(6) which is divorce), SSA PIA, Wealthbox domicile, current CMS IRMAA thresholds.
• Five conversion window types: pre-RMD (age 55-72 before SECURE 2.0 trigger), pre-Social-Security (62-70 delay window), low-income year (sabbatical / job transition / sale / recovery), post-business-sale (coordinated with L3 Ch6 L4 QSBS / installment / CRT / opportunity zone), inherited-IRA coordination (L3 Ch3 L2 10-year-rule).
• Leverage ratio = long-term tax saved / current-year tax paid. Components: current-year cost (federal bracket fill + NIIT + state + IRMAA cliff + pro-rata) vs lifetime savings (reduced RMDs + surviving-spouse single-filer + bequeath beneficiary + lifetime IRMAA + lifetime state). Hendersons example: leverage 2.4 on $74K conversion.
• Three edge cases: negative leverage (defer with reason), cliff-leverage (optimal conversion at the cliff edge not bracket-fill max), coordinated leverage (joint optimization with NUA / 409A / charitable / QCD).
• The prioritized screen output is a ranked table per L2 Ch8 L2 JSON schema (household_id, ages, AGI, window type, recommended size, current cost, long-term savings, leverage ratio, flags, pass-state, next-action), tagged-Markdown-routed to Smarsh, reviewed and authorized by senior advisor for L3 Ch2 L2 advancement.
• Recovered hours typical 80-130 per quarter for a 200-household practice. The screen converts ad-hoc memory-driven conversation into systematic evidence-driven workflow, surfaces the missed 30+ household cohort, and feeds the L3 Ch2 L2 sizing memo and L3 Ch2 L3 Reg BI recommendation/execution/archiving workflows.