Methodology Overview

The Lifetime Tax Planner is a deterministic, rule-based lifetime simulation: every number traces to a published tax rule, a documented assumption, or an input you control. The same inputs on the same engine version and simulation settings always produce the same outputs; an update can legitimately change a result, and the release notes say when. Calculations come from a deterministic rules engine — not a machine-learning model. Projections depend on your inputs and the disclosed economic, longevity, and healthcare assumptions; actual outcomes will differ. This page states what is modeled, how the optimizer chooses a strategy, what is deliberately excluded, and how the engine is validated.

What the engine models

• Federal income tax — official 2026 brackets and standard deduction as published (Rev. Proc. 2025-32, post-OBBBA), inflation-indexed beyond 2026; 65+ additional deductions including the OBBBA senior bonus with its phase-out and expiry; capital-gains 0/15/20% stacking per the IRS Qualified Dividends & Capital Gain worksheet; NIIT; the Child Tax Credit (nonrefundable portion — the refundable additional credit is not modeled) and Credit for Other Dependents; FICA in working years.
• Social Security — taxability per the IRS Pub 915 worksheet; claiming-age reduction and delayed-retirement credits; spousal benefits at household level; the claiming age is an optimized decision variable.
• State income tax — all 50 states + DC from Tax Foundation 2025 data: progressive brackets and standard deductions where they exist (uniformly inflation-indexed as an approximation — not every state indexes its thresholds; a per-state indexing audit requires statute-level verification per state and is scheduled with the January 2027 table review; filing-status aware), whether the state taxes Social Security, and retirement-income exemptions/exclusions (e.g., PA/IL/MS/IA-style full exemptions; GA/NJ/CO-style 65+ exclusions).
• Medicare IRMAA — tier thresholds and surcharges per SSA POMS, applied with the statutory two-year MAGI lookback; conversions are optionally capped below tier cliffs.
• ACA premium credits — IRC §36B expected-contribution schedule with the 400%-FPL cliff (returning 2026 under current law); pre-65 conversion ceilings respect the cliff.
• Retirement accounts — 401(k)/403(b)/457(b) (separate limits), traditional/Roth/SEP/SIMPLE IRAs, HSA, taxable lots with basis tracking, 529s (per-child phases: private school, college, grad school), NQDC, cash-value life. Contribution limits and catch-ups (including 60–63) clamp everything; employer match formulas feed §415(c) room.
• Early access before 59½ — Rule of 55, SEPP/72(t), 457(b) separation access, Roth basis, HSA-with-receipts, 529 basis, and the 10% penalty where it truly applies (§72(t), §223(f), §529(c)).
• RMDs — Uniform Lifetime Table with SECURE 2.0 start ages (73/75) by birth year.
• Strategies — Roth conversion schedules that respect bracket, ACA, and IRMAA limits, gains harvesting in 0% years, backdoor and mega-backdoor Roth (pro-rata modeled), QCDs, 529→Roth rollovers (per-beneficiary $35k streams), securities-backed lines of credit including buy-borrow-die mode with margin-call modeling, annuitization, home downsizing/moves (§121), international retirement (residence-based pieces switch; US federal follows), survivor analysis (single filing after a first death), long-term care costs, and a de-risk return step.
• Risk analysis — up to 13 named stress scenarios (some apply only when they can: a working spell before retirement, entered Social Security, a securities-backed line, a move abroad), re-run beside the base case (a failing scenario opens to its first unfunded year, worst single-year gap, total unmet spending, and three one-change adjustments re-run under the same scenario; the strategy is never re-optimized); seeded Monte Carlo with per-year normally-distributed return and inflation draws (reproducible, not cherry-picked; 250 draws by default, adjustable). Success percentages carry sampling uncertainty — roughly ±2 points near the extremes and up to ±6 points near 50% at 250 draws — and are displayed with that margin: they are illustrative, not precise. Normal draws underrepresent severe market events; the deterministic stress scenarios exist to cover exactly that tail. One-at-a-time sensitivity ranking completes the picture.

How optimization works

The planner evaluates multiple tax, claiming, and contribution strategies across the household's projected lifetime, simulating each one through the full engine. It prioritizes keeping the plan funded, preserving necessary liquidity, and then improving the user-selected objective (balanced, minimize taxes, minimize penalties, minimize IRMAA, maximize robustness). The entered plan is always evaluated alongside the alternatives, so the chosen strategy never funds worse than doing nothing. Results represent the best strategy among those modeled — not proof of a globally optimal or individually suitable recommendation. When a chosen strategy accepts more lifetime tax in exchange for funding, wealth, or robustness, the app says so explicitly, and structural suggestions are always presented as recommendations with their trade-offs stated — never applied automatically. Sustainable-spending capacity is a derived diagnostic computed under the chosen strategy, not a selection objective.

What is deliberately excluded

This is planning software, not tax preparation. Not modeled, by design and disclosed in-app: AMT, tax credits beyond the child, other-dependent, dependent-care and ACA premium credits, county/city income taxes (MD counties, NYC, OH municipalities), Utah's credit-based system, Washington's capital-gains excise, the SIMPLE-IRA two-year rule, SEPP payment-method precision, treaty-specific international rules, FEIE, the §877A exit tax, state estate and inheritance taxes, and Medicaid planning. Federal estate tax IS applied to the net-to-heirs figure above the inflation-indexed exclusion, and securities-backed lines carry maintenance margin calls with forced sales. Monte Carlo models market randomness, not every real-world risk. Where a simplification could matter to you, the in-app Audit page (under Results) lists it next to the assumption it affects.

Two structural choices worth knowing: working years are contribution-driven, not cash-flow-simulated — the accounts grow by returns plus your entered contributions, and income above taxes, contributions, and debt payments is treated as spent rather than auto-invested, so changing income updates taxes and contribution limits but not the contributions themselves. Warnings cover both failure directions (contributions income can't fund; a very large unallocated surplus), and an optional savings rate (% of income) makes taxable saving scale with each spouse's actual wages for households who save their surplus. And on ACA subsidies: when no benchmark premium is entered, the pre-65 healthcare amount (or its entered premium portion) stands in as the benchmark — entering the premium portion caps the modeled credit at premiums actually paid (out-of-pocket costs are never subsidized), and entering the real second-lowest-cost silver premium gives the most accurate subsidy math.

How it is validated

• 1564 automated tests run on every change (count and engine version v2.326 are generated from the current release at publish time; also pinned in-app on the Why Trust This tab).
• A hand-calculated validation library: tax cases computed by hand from IRS, SSA, CMS, and HHS primary sources with every intermediate step shown, plus a fully hand-calculated multi-year withdrawal schedule — the engine matches each case exactly (schedule years within a dollar), and every case is a permanent automated test. Two cases were additionally hand-verified by an independent external reviewer.
• The two highest-risk calculations — Pub 915 Social Security taxability and the Qualified Dividends & Capital Gain Tax worksheet — are cross-checked against independent recomputations of the IRS worksheets across a large sweep of randomized taxpayer situations.
• Hand-verified calculations are pinned by permanent tests: bracket math, state cases (CA/NY/OH/MA/NC), IRMAA tiers, ACA schedules, contribution-limit clamps, SS claiming factors, amortization, §121.
• Money is conserved in every simulated year: sources + gap = spending + taxes + penalties, asserted automatically.
• Every recommendation in the app carries an ⓘ that names the governing rule and how that specific logic is tested.

Rule-by-rule validation

Every modeled rule, its cited source (linked), the effective year of the seeded values, and how that rule is tested. Links point at primary sources wherever one exists — the IRS, CMS, SSA, the Federal Register. Two are deliberately not primary and say so: state data comes from the Tax Foundation's dataset (a secondary compilation), and statute links use Cornell's Legal Information Institute (a reliable reproduction of the U.S. Code, not the official publisher).

Data vintage and review cadence

Federal tables are the official tax-year 2026 values as published — IRS Rev. Proc. 2025-32 (brackets, deductions, capital gains), IRS Notice 2025-67 (retirement plan limits), CMS (2026 IRMAA tiers), SSA (wage base) — not reconstructed by inflating a prior year. State tables are Tax Foundation 2025. Future years are projected by inflation-indexing from these official values — projected, not enacted law. Tables are re-verified annually, not daily; the next scheduled review is January 2027 (2027 IRS inflation adjustments), with corrections recorded in the public release history. Strategies whose payoff depends on a specific statute surviving (the §1014 step-up, backdoor conversions, SECURE 2.0 rollovers) carry law-change warnings at the point of use.

Free educational tool — not tax, legal, or investment advice. See the Disclaimer. Questions about the methodology: [email protected].