Editorial methodology, sources & review process

Who reviews the numbers

Tax inputs and calculator methodology are reviewed by Deepak Middha, Chartered Accountant, who has 18 years in hedge-fund accounting and taxation and has passed the Series 65 Uniform Investment Adviser Law Examination. He is the founder of TaxSaveIQ and its tax calculator methodology reviewer.

Passing the Series 65 examination is not a licence or registration, and TaxSaveIQ does not provide investment-advisory services. Review means the tax data, formulas, and assumptions are checked against official sources — it is not an audit, and it is not personal advice.

What a calculator review covers

• Tax-year inputs and which year each calculator defaults to
• Federal brackets and threshold figures
• Standard deduction amounts by filing status
• Credits, phaseouts, and their income ranges
• Retirement contribution and catch-up limits
• Capital-gains rate thresholds
• Self-employment tax methodology, including the 92.35% adjustment and the deductible half
• Calculator formulas and the order in which deductions are applied
• Stated assumptions and what each tool deliberately leaves out
• Edge-case handling, such as phaseout boundaries and wage-base limits
• Source links to the governing official material
• Plain-English explanations of the results
• The last-verified date shown on the page

How sources are ranked

Different calculators rely on different official material. When two sources appear to disagree, the higher tier wins:

Law, proposed law, and IRS guidance are not the same thing

These three get conflated constantly in tax coverage, and the difference decides what a calculator is allowed to show.

How a tax year’s data gets onto the site

A tax year’s figures do not arrive all at once. They arrive in stages, and each stage changes what a calculator is allowed to display.

What official, enacted, preliminary and fallback mean

Every federal figure the calculators use is stored with a source URL, that source’s publication date, the date it was last verified, and one of four status values. Of the 52 federal figures currently in the 2026 data set, 23 are official, 29 are enacted, 0 are preliminary and 0 are fallbacks.

The 2026 federal data set was last verified 2026-07-28.

Which tax years the calculators cover

The calculators compute 2025 and 2026. 2026is the default. Each year’s figures live in their own data file with their own sources and verified dates, and one year’s constants are loaded per calculation — so a 2025 figure cannot appear in a 2026 answer, or the other way round.

Asking for any other year produces a clear error rather than an answer. A calculator that quietly replies to a 2031 question with 2026 rates is worse than one that says it does not know: the first looks like an answer.

2025is kept because the law changed retroactively. The One Big Beautiful Bill Act raised that year’s standard deduction, Child Tax Credit and state-and-local tax limit after the original inflation adjustments were published, so a 2025 return does not match the figures released in late 2024. Long-range projections, such as the lifetime calculator, apply current law to future years and say so — they do not guess at unpublished inflation adjustments.

The order the calculation runs in

Order matters as much as the rates do. A deduction taken at the wrong step produces the wrong answer even with perfect figures — which is why gross income, adjusted gross income, modified adjusted gross income and taxable income are four separate values in the engine rather than one field called “income”. Every federal figure on the site comes from one shared calculation, run in this sequence:

1. Gross income — wages, business profit after business expenses, capital gains, dividends, interest and other ordinary income.
2. Above-the-line adjustments — retirement and health-account contributions, the deductible half of self-employment tax, and the other Schedule 1 Part II items.
3. Adjusted gross income.
4. Modified adjusted gross income, computed separately for each provision that needs one rather than reusing a single number.
5. The standard deduction, including the additions for age 65 and over and for blindness, or itemized deductions if they are larger.
6. Schedule 1-A deductions for qualified tips, qualified overtime, car loan interest and the senior deduction.
7. The charitable deduction available to taxpayers who do not itemize.
8. The section 199A qualified business income deduction.
9. Taxable income.
10. Splitting taxable income into the part taxed at ordinary rates and the part taxed at the 0%, 15% and 20% capital-gain rates.
11. Ordinary income tax from the rate schedule for the filing status.
12. Capital gain tax, stacked on top of ordinary income as the statute requires.
13. Nonrefundable credits, then the refundable Additional Child Tax Credit.
14. Social Security tax, Medicare tax, Additional Medicare Tax and self-employment tax, sharing one Social Security wage base.
15. The 3.8% Net Investment Income Tax.
16. Refund or balance due, after withholding and estimated payments.

Because there is one implementation, the same inputs give the same federal answer on every calculator, worked example and comparison page on the site. This list is rendered from the engine itself, so it cannot drift from what the code does.

Rounding

Small rounding choices compound. A return involves twenty or more additions, and floating-point dollars drift. The policy is therefore:

• Every dollar amount is converted to a whole number of cents as soon as it enters the calculation, rounding half away from zero.
• All intermediate arithmetic is done in whole cents, so no fractional-cent drift can accumulate across the twenty-odd additions in a return.
• When a rate is applied to an amount, the product is rounded to the nearest cent immediately.
• A progressive rate schedule is summed once and rounded once at the end, rather than rounding each bracket separately.
• Rates such as the marginal and effective rate are returned unrounded and are rounded only for display.
• Amounts are rounded to whole dollars only at the moment they are shown on screen. The underlying figure keeps its cents.

A displayed figure can therefore differ from your own arithmetic by up to fifty cents, because it is a rounded view of an exact cent amount. The IRS itself allows returns to be completed in whole dollars.

Source map by calculator

Each calculator cites the material that actually governs its calculation. These are the same sources listed in the “Methodology and sources” section on each page.

Other tools — the W-4 optimizer, RMD calculator, Roth conversion calculator, tax-loss-harvesting calculator, quarterly planner, and the OBBBA and state pages — cite their own sources and verified dates on the page itself.

Verification cadence and dates

• Federal figures are re-checked when the IRS publishes new inflation adjustments or new guidance on a provision we model.
• State figures are re-checked per state; where a state has not published 2026 amounts, the latest enacted figures are used and the page says so.
• Each calculator and guide displays the date its tax data was last verified, taken from the same data file that drives the calculation.
• A page's last-verified date changes only when the underlying tax data is re-checked — not when wording or layout is edited.

We publish two separate dates, because they mean different things. A methodology review is a human re-verification of the data, the formulas, the official sources, the worked examples, the FAQ answers, the structured data, and the edge cases. A content updateis a change to wording, layout, or links. A copy edit never moves the methodology date — otherwise “reviewed” would stop meaning anything.

“Last verified” is narrower than either. It is the date a person opened the cited official document and confirmed that the number on this site is still the number in that document. It is not the date the page was edited, not the date it was deployed, and not the date the source was published. If a figure has never been checked against a current-year source, it does not get a verified date — it stays a fallback and cannot be displayed as a final figure at all.

Review log

Every review date shown anywhere on the site — on a calculator badge, in a “Methodology and sources” section, and in each page’s structured data — is rendered from this one record, so no two surfaces can disagree.

• federal-tax: Re-reviewed on 2026-07-25 because the business-mileage input feeds this calculator and the 2026 rates changed on July 1. The mileage input is now split into two periods at the published rates; brackets, the standard deduction, OBBBA deductions, QBI and the FICA/SE logic were re-checked against Rev. Proc. 2025-32 and are unchanged.
• state-tax: Full re-review on 2026-07-25. The federal SALT panel now applies the statutory 2026 limitation ($40,400 / $20,200 MFS, phased down above $505,000 / $252,500 MAGI at 30% of the excess, floored at $10,000 / $5,000) from lib/tax/federal2026.ts instead of a flat cap. Corrected a double-counted resident-credit in the multi-state total, and re-scoped the live/work-state feature as guidance rather than a reciprocity computation.
• self-employment-tax: Batch 3 closeout on 2026-08-01: rebuilt on the shared federal engine, corrected the Social Security wage-base coordination with W-2 wages (Schedule SE lines 8a-8d), and implemented the section 199A W-2 wage limit, its phase-in, the SSTB rules and the OBBBA minimum deduction. W-2 wages and qualified-property UBIA are collected rather than assumed.
• w2-vs-1099: Batch 3 closeout on 2026-08-01: rebuilt as a break-even rate solver with two separated models, spendable cash and total economic value. Corrected a double-count that charged the same benefit gap to both sides and overstated the break-even rate by up to $10.63 an hour.
• llc-tax: Batch 3 closeout on 2026-08-01: entity-level state taxes modelled for California, Texas and Illinois; every payroll, entity and operating cost separated; baseline relabelled a single-member LLC taxed as a sole proprietor because it is charged the LLC annual tax; the section 199A wage limit applied on both sides.
• roth-vs-traditional: Re-verified on 2026-07-29 in the Batch 2 production closeout: the model in use is stated above the recommendation, the taxable side account is shown as its own line on both comparisons, and the displayed Traditional total is checked programmatically against its own line items to within one cent. The existing-IRA-balance field is now named for what it is and no longer reads as a seed balance that grows.
• lifetime-tax-savings: Re-verified on 2026-07-29 in the Batch 2 production closeout: the annual ledger now carries spouse age, filing status, beginning cash, wages and pension split apart, the deduction applied, and federal ordinary tax, capital-gains tax and NIIT as separate columns; the CSV export carries the scenario, generation timestamp, assumptions and tax-data version in its header. State tax and IRMAA are described in one place and imported everywhere. A survivor transition from joint to single filing is an explicit input and is reported per year in the ledger.
• family-tax-credit-finder: Batch 5 on 2026-08-03: the Earned Income Credit is now computed on the IRC § 32 curve rather than reported as a yes/no flag, and two 2025 figures were replaced with their 2026 values — the disqualifying investment-income limit ($11,950 → $12,200) and the completed-phaseout ceilings. The Child and Dependent Care Credit moved to the OBBBA § 70404 rate schedule, whose top rate is 50% rather than the pre-OBBBA 35%, with a second phase-down above $75,000 ($150,000 joint). The adoption credit gained its 2026 refundable portion of $5,120, and the dependent-care FSA offset against the credit's expense limit was implemented.
• home-office-deduction-calculator: Batch 3 closeout on 2026-08-01: inputs regrouped into direct, indirect, internet and phone, depreciation and simplified, each taking exactly one allocation. Corrected a defect that applied the square-footage percentage to internet after its own business-use percentage had already been applied.
• irs-form-wizard: Batch 5 on 2026-08-03: rebuilt so every form is produced by a predicate over an actual answer, and split into forms the answers establish and forms that may apply, each carrying its reason and its deciding condition. Form 8995 became conditional and now names Form 8995-A; Form 1040-ES is no longer pushed at a W-2 employee with small investment income; Form 2441 keys off work-related care rather than merely paying for care. Added crypto, rental property, marketplace insurance, RMDs, Roth conversions, foreign accounts and gifts, and a published list of what the wizard does not cover.
• mileage-deduction-calculator: Batch 3 closeout on 2026-08-01: added the actual-expense method, parking and tolls as a separate line, first-year election and multiple-vehicle warnings, and replaced the flat 14.1% self-employment-tax claim with a real marginal payroll-tax calculation that accounts for the Social Security wage base.
• quarterly-estimated-tax-planner: Batch 5 on 2026-08-03: the schedule is now date-aware. It reconstructs the Form 2210 ledger from payment dates and amounts, marks each installment paid, underpaid, past due, due next or future, and never presents a due date that has passed as a payment to make. Added the annualized income installment method, including the prorated Social Security wage base in Schedule AI Part II, the ratable withholding rule and the Form 2210 box D election, and an explicit warning that equal division is not penalty-safe after a missed installment.
• roth-conversion-calculator: Re-verified on 2026-07-29 in the Batch 2 production closeout: bracket room is now reported as four separate values — total room before the conversion, room used, room remaining after it, and the amount spilling into the next bracket — measured against taxable income, with the base standard deduction and the IRC § 63(f) age or blindness addition shown on their own lines. The worked example is generated from the same function as the calculator and matches its default inputs exactly.
• tax-refund-estimator: Batch 5 on 2026-08-03: rebuilt on the shared federal engine and laid out as the Form 1040 settlement — total tax on line 24, then payments, then the difference. Self-employment tax now reaches the result, which the previous standalone calculation could not express, so a contractor with no withholding is no longer shown a refund. Nonrefundable and refundable credits are separated, and the refund is split into the part that is over-withholding and the part that is genuinely a credit.
• tax-savings-action-plan: Re-reviewed on 2026-07-25: the mileage recommendation now states both 2026 rate periods instead of a single full-year rate. Scoring and every other move were re-checked and are unchanged.
• w4-withholding-optimizer: Batch 5 on 2026-08-03: replaced the year-to-date pace projection with the employer's own calculation — the Publication 15-T (2026) percentage method, whose derived rate schedules are verified row by row against the 48 rows the IRS printed. The tool now solves for the Step 4(c) figure that reaches a chosen target over the paychecks actually remaining, and models Step 2 treatment, Step 3 credits and their income limit, Steps 4(a) and 4(b), other jobs and a working spouse, a bonus at the 22% supplemental rate, and self-employment tax.

How a figure is cross-checked before it ships

Agreeing with another calculator is not evidence of being right — two tools can be wrong in the same way, and often are, because they copy each other. Every changed calculation goes through the same sequence:

1. The expected answer is first worked out by hand from the statute and the IRS rate schedule — never by copying another calculator.
2. That hand calculation becomes a unit test, so the figure is re-checked on every commit rather than once.
3. The same scenario is then run through at least two reputable independent calculators that support the same tax year and the same assumptions.
4. Filing status, tax year, income definition, deductions and credits are matched exactly before any comparison is made. Most apparent disagreements between calculators are really different assumptions.
5. Any unexplained difference above $2, or 0.1% of the figure if that is larger, is treated as a failed test and investigated before the change ships.
6. Where the official formula supports our result, our result stands. A third-party tool is a cross-check, not an authority, and we do not change a calculation merely to match one.

What the automated tests cover

A human review happens on a schedule; the tests run on every change. They are what stops a correct figure from being quietly broken by an unrelated edit months later. The suite covers, at minimum:

• Every filing status, at zero income and at high income.
• Each federal bracket boundary at one dollar below the threshold, at the threshold, and one dollar above it.
• The Social Security wage base boundary, and the coordination of that single annual cap across wages and self-employment earnings.
• The Additional Medicare Tax thresholds, measured against combined wages and self-employment earnings rather than each source separately.
• Child Tax Credit phase-out boundaries, and the point at which the credit reaches zero.
• The capital-gain rate boundaries, including the stacking of gains on top of ordinary income.
• The Net Investment Income Tax threshold boundaries.
• The One Big Beautiful Bill Act deduction phase-outs.
• Negative business income, empty inputs, and non-numeric inputs.
• That the same inputs produce the same federal figures on every page that uses the shared engine.

Boundary cases get three tests each — one dollar below a threshold, exactly at it, and one dollar above — because off-by-one errors at a threshold are the most common way a tax calculation goes wrong while still looking plausible.

What the state and local estimates can and cannot do

The federal calculation is modelled in full. State and local tax are deliberately held to a lower claim, and each page says which it is making.

• State figures come from each state’s own revenue department and carry their own verified date. They are never derived from the federal figures.
• A state page is labelled a full calculation only where the rates, brackets, standard deduction, exemptions and the treatment of retirement income are all modelled. Anything less is labelled a simplified estimate, and the label says what is simplified.
• Local and municipal income tax — Ohio cities, Maryland counties, New York City, Pennsylvania localities and others — is only included when you select a jurisdiction. Without one, the page says local tax is excluded rather than quietly returning zero.
• Credits, deductions and add-backs peculiar to one state are generally not modelled. Where a state has not published its 2026 figures, the latest enacted figures are used and the page says so.
• Non-income state levies that still come out of a paycheck — Washington’s Paid Family and Medical Leave and WA Cares premiums, for instance — are shown as separate lines and never folded into an income-tax number.
• Reciprocity between states, part-year residency and multi-state allocation are not modelled at all. If you moved or worked across a state line, treat the state figure as indicative only.

Corrections policy

• If a figure is wrong, we fix the underlying data rather than patching a single page, so every calculator that uses it is corrected at once.
• Corrections that change a calculated result also update that page's verified date.
• Where the law is unsettled or a state has not published 2026 figures, the page says so instead of presenting an estimate as final.
• Found something wrong? Report a correction — include the page and the figure you believe is incorrect.

What TaxSaveIQ does not do

• No individualized tax, legal, accounting, financial, or investment advice.
• No tax return preparation, e-filing, or representation before the IRS.
• No claim of IRS approval, certification, or endorsement — TaxSaveIQ cites IRS material, which is not the same as being endorsed by the IRS.
• No user ratings or reviews are collected, so none are published or marked up in structured data.