How we check taxes, health costs, and retirement income

The short answer

A retirement model needs to do more than draw market returns. It must also work out how much leaves your savings to pay bills, health costs, and taxes. Those parts can affect one another.

We reran our engine's financial benchmark suite on September 11, 2026. All 607 checks passed. They include 37 tax cases compared with saved outputs from outside tax software. We also checked four made-up households across 51 years against a separate calculation of the same plans.

The checked tax amounts agreed within one cent. The checked household dollar amounts agreed within two cents. This is evidence for the math in these cases, not a claim that every rule or future event is covered.

The 607 checks are software tests, not 607 separate households or outside reviews. Some check the same rule in several ways.

This page tests engine features. Not all are enabled in the current app. In particular, these tests turn on detailed health-insurance credits and Medicare income surcharges that the current When You Stop app does not use.

What is in the app today?

An engine feature can pass a test before a product lets users choose it. We keep those two claims separate.

A preview can show sample results instead of a live engine run. The benchmark findings below come from the real Python engine, not those sample screens. See our full methodology for the app's other assumptions.

How the yearly pieces fit together

Think of each year as a set of accounts to balance. In these test cases, we take spending out before applying that year's return.

1. Set the year's spending and health-cost targets. Apply the chosen price-change rules.
2. Work out Social Security and any required account withdrawals.
3. Find how much more cash must come from savings. The tax cost depends on which account pays it.
4. Recheck taxes and any enabled health-insurance rules at that income. A larger taxable withdrawal can itself create more tax.
5. Settle the bills, track transfers between accounts, and put any spare cash back into savings. Then track the year's returns and closing balances.

This is a plain-English summary, not a claim that every rule is calculated just once in this order. Some amounts must be solved together until the cash and tax figures agree.

A Roth conversion moves money from a pretax account into a Roth account. The transfer is not lifestyle spending. But it can raise taxable income, reduce an enabled health-insurance credit, and affect a later Medicare surcharge. A tax payment is money leaving the portfolio; moving money between its accounts is not.

What we compare against

Different checks offer different kinds of evidence. A test written by us is not the same as a result from another tool.

The tax reference uses PSLmodels Tax-Calculator with its own policy data, not Whatify's tax brackets. It checks deductions, capital gains, taxable Social Security, and selected income cutoffs. The policy basis includes IRS guidance for 2026. These are rate calculations, not completed tax filings. State tax, payroll tax, itemized deductions, and several other taxes and credits are outside this comparison.

The health checks use saved KFF calculator results, IRS contribution rules, and CMS surcharge tables. ACA means the Affordable Care Act. An ACA credit can lower eligible insurance premiums before Medicare. IRMAA is an extra Medicare charge tied to income, usually from two years earlier.

For income, the checks use SSA claim-age rules and survivor rules. Required minimum distributions, or RMDs, use the relevant IRS tables. A published-rule match tests our reading and coding of a rule; it is not an endorsement by that agency.

Four households that test the links between rules

All four cases start with $2 million in a pretax retirement account, no taxable or Roth savings, and $45,000 of yearly lifestyle spending in opening-year buying power. A supplied $12,000 health-premium budget is extra. After age 65, that budget includes standard Medicare costs; modeled income surcharges are added separately.

Real market returns, market swings, and fees are set to zero. This keeps random market luck out of the accounting check. Each household follows one fixed path; repeated identical paths would not add new evidence. With 2% inflation, zero real return means 2% nominal return. Cash flows use that year's dollars; real ending balances use opening-year buying power.

We wrote the household reference separately, using outside tax software and published rules. It did not import our engine, use its policy files, or copy its answers. Still, both calculations could share a mistaken reading of a rule. This is not an independent professional audit.

A conversion can affect more than one year

In the single-person case, year two includes a $120,000 Roth conversion. It removes that year's ACA credit in this test and raises federal tax. Two years later, the income from that year creates a Medicare surcharge. Social Security starts at 67. Required withdrawals start later and can leave more cash than the household needs to spend.

Here are selected amounts from the no-inflation case. The reference and engine columns are rounded to cents. These are measured test results, not typical costs or advice to make a conversion.

The CSV download gives tax, net health costs, Social Security, required withdrawals, and ending real savings for all 51 years, not just these selected events. The largest difference across all checked household dollar fields was less than one cent. Our fixed limit was two cents, with no percentage-based allowance.

What the tests do not prove

• The four households are well funded. This separate reference does not test what happens when every account runs out. We have internal shortfall checks, but that is a different kind of evidence.
• Supplied health premiums are inputs, not forecasts of medical bills, long-term care, or local insurance prices.
• Taxes and health costs follow stated annual rules. The cases do not cover every filing choice, coverage decision, death-month rule, or appeal.
• The current app does not enable every tested engine option. Turning those options on needs its own product and end-to-end checks.
• Correct yearly math does not prove a return forecast or a retirement success rate. Our market-method guide answers that separate question.

The tests also try a deliberate $1 tax error. The household comparison catches it. We do not repair a failed math check by lowering expected returns, changing a seed, or widening the allowed difference.

Dates, versions, and how to check the findings

This rerun uses engine 0.15.0, result format m16.0, and saved reference outputs. The tax-software and separate household references were generated on September 6, 2026. The KFF cases and CMS tables were recorded on August 11, 2026. A new test date does not make an older source new.

We did not rerun the outside calculators or certify the latest tax-software release for this page. The tax reference stays pinned to version 6.5.3. The run details include exact household inputs, versions, source fingerprints, tolerances, measured differences, and the test-suite result. The engine is private, so these downloads help readers inspect our claims but do not provide full outside reproduction.

Version 1.0 is the reviewed publication edition. Its measured results are unchanged from the initial study. Future updates should add tougher account-exhaustion cases, review newer outside software, and compare complete household plans with another planner or a qualified reviewer. Those are next steps, not completed validation.

This guide explains tested calculations. It is not financial, tax, insurance, or legal advice.