Methodology

Editorial principles

CalcBold is a free calculator hub. The thing that distinguishes it from a thousand other calculator sites is not the count and not the design — it’s that we tell you, on every single page, exactly which formula produced your number and where that formula came from. If you can’t verify a calculator’s math against a primary source, the calculator should not be on the open web. That is the bar.

Primary sources only. Every formula on this site traces back to a government dataset, a peer-reviewed study, a standards-body document, or an industry specification published by the entity that owns the math. We do not use Wikipedia as a citation, we do not use other calculator websites as a citation, and we do not cite generic SEO blogs. When the only available source for a number is an aggregator summary, we flag it on the page and treat the calculator as opinion-grade rather than authority-grade.

Formula transparency.Every calculator’s page renders the actual formula it ran — not a prettified marketing version, the literal math the JavaScript executes. If the result depends on a piecewise function (tax brackets, insurance tiers, BMI bands), the page lists every breakpoint. If the result depends on a default that the calculator picked for you (a discount rate, an inflation assumption, an emissions factor), the page surfaces that default and lets you change it. There is no hidden coefficient on this site.

Annual review.Every calculator gets at minimum one source-and-default sweep per calendar year, anchored to the publication cycle of its underlying authority (IRS Rev. Proc. drops in October-November; HMRC PAYE in March-April; CBDT in February). Some calculators get more frequent reviews — AI vendor pricing quarterly, fuel and electricity tariff defaults quarterly, cost-of-living monthly. The footer of every calculator page stamps the snapshot date so you know how stale the defaults are when you load the page.

Error correction. When a reader emails [email protected] with a math bug, the triage rule is: if the bug changes the headline number by more than 1%, we treat it as a P0 incident and ship a fix the same day. Smaller drift goes into the next weekly queue. Every fix that touches a published formula gets a corrections note added to the page footer with the date and one-sentence summary of what changed.

Per-domain methodology

CalcBold ships 220 calculators across 15categories. The rest of this page works through the major domains in turn — the authority whose formula we use, the dataset we calibrate defaults against, and the claims we explicitly do not make.

Tax and deductions

For US federal calculations we use the bracket tables, standard deduction, FICA ceilings, AMT exemptions, and contribution limits published in the relevant IRS Revenue Procedure (Rev. Proc. 2024-40 governs tax year 2026 inflation-adjusted amounts). For UK calculators we use HMRC PAYE (income tax bands, National Insurance categories, student loan thresholds for the current 2025/26 tax year). For India we use the income tax slabs and surcharge structure published by the Central Board of Direct Taxes for the relevant assessment year.

What we explicitly don’t do: state-level US tax (only six states are modeled and each is flagged on its calc), provincial tax in Canada, or local council tax in the UK. Where multiple jurisdictions could plausibly apply (RSU vesting across state lines, bonus withholding) the page warns about the limitation rather than guessing. We also don’t model phaseouts that depend on circumstances we can’t infer from the form (e.g. the SALT cap interaction with the AMT in edge cases) — those calculators say so explicitly and recommend a CPA.

Finance

Loan, mortgage, and amortization math uses the standard reducing-balance formula documented by the CFPB and matched against the underwriting math in the Fannie Mae Selling Guide and Freddie Mac Single-Family Seller/Servicer Guide. Risk-free benchmark rates for spread comparisons (CD vs T-bill, savings vs money market, real-yield calcs) come from the Federal Reserve H.15 Selected Interest Rates release, refreshed monthly. Retirement contribution limits, RMDs, and catch-up provisions follow IRS Pub. 590-A/B for the active tax year.

EMI = P × R × (1 + R)^N / ((1 + R)^N − 1)— that exact equation is what every loan calc on this site runs. We don’t use the “flat rate” shortcut some lender pages quote, because flat-rate interest is a different product and conflating them is how borrowers end up with surprise balances. Where a calculator includes inflation adjustment we note whether the displayed number is nominal (today’s dollars) or real (purchasing-power adjusted) and let you toggle between them.

Health and longevity

BMI uses the WHO BMI band definitions (underweight <18.5, healthy 18.5–24.9, overweight 25–29.9, obese ≥30) without modification. Resting metabolic rate uses the Mifflin-St Jeor equation (1990) because it predicts measured RMR more accurately than Harris-Benedict in modern, non-athletic populations. Activity multipliers for TDEE come from the Compendium of Physical Activities (Ainsworth et al.) MET tables. Pregnancy due-date math uses the Naegele rule recommended by ACOG Committee Opinion 700.

What we explicitly don’t do: medical diagnosis. Every health calc carries a disclaimer that BMI is a population-level screening tool, not a diagnosis of individual body composition; that TDEE is an estimate that real-world adherence biases by 10–20%; that biological-age calcs are research-grade approximations, not clinical assays. The whitelist of population groups for which Mifflin-St Jeor was validated does not include children under 18 or competitive athletes; those calcs say so.

Climate and energy

Solar generation modeling uses the methodology behind NREL PVWatts (TMY3 weather data, DC-to-AC derating, tilt and azimuth corrections), with default panel degradation rates calibrated against NREL/LBNL field-study median values. Fuel pricing defaults come from the EIA weekly retail price series. Carbon-intensity factors for electricity, fuel, and aviation come from the EPA Emission Factors Hub and ICAO methodology for flight-segment estimates. Heat-pump COP defaults and insulation R-value tables are anchored to ENERGY STAR certified-product baselines.

What we explicitly don’t do: utility-specific tariff modeling beyond a flat average price (real-world rate plans with TOU, demand charges, and net-metering buyback caps require utility-specific data we don’t have). Solar payback math uses straight-line ROI, not a discounted cash-flow model with stochastic price forecasting; the calculator notes the limitation.

AI and tech

Token cost, agent economics, and inference budget calcs use the published list prices from each vendor’s pricing page — Anthropic, OpenAI, Google AI, Mistral — refreshed quarterly. Context-window limits come from each vendor’s API documentation (not third-party leaderboards). Where a calculator compares model quality, the source is HELM or MMLU scores, which are flawed but at least public and reproducible.

What we explicitly don’t do: predict that vendor X will still be cheaper in 12 months. The list-price reality at this layer of the market is that pricing changes quarterly. Every AI calc shows the snapshot date and warns that a re-run after the next round of vendor updates may give different numbers.

Career and business

Salary benchmarks, geographic-arbitrage adjustments, and freelance-rate floors anchor to BLS Occupational Employment and Wage Statistics (OEWS, formerly OES) median wage tables by occupation and metro. We cross-reference against self-reported aggregators (Glassdoor, Levels.fyi, Built In) to surface outliers and stack-rank tech-specific roles where BLS occupational categories are too coarse. The calculator surfaces both numbers when they disagree by more than 15% and asks you to pick which one fits your situation rather than averaging them silently.

What we explicitly don’t do: claim that the “market rate” for any single negotiation is the median number we show. Real offers depend on tenure, scope, counter-offer strength, and the hiring manager’s budget envelope — none of which a calculator can know. The output is a benchmark band, not a target. Self-reported aggregator data (Glassdoor, Levels.fyi) carries a survivorship-bias warning on every page that uses it: people who disclose tend to be people who got paid well, so treat the upper percentiles as aspirational rather than typical.

Decision and life

Cost-of-raising-a-child uses the USDA Expenditures on Children by Families report as the core dataset, with adjustments for the post-2017 cost environment based on BLS CPI sub-indices (food at home, childcare, transportation). Behavioral-economics primitives used in regret-discount, decision-fatigue, and probability-of-success calcs draw on the published research literature (cited per-calc) — we lean on primary papers from Stanford, Yale, and the National Bureau of Economic Research rather than secondhand summaries.

What we explicitly don’t do: predict your individual outcome. These calcs translate a research base rate into a personalized framing — they do not forecast your specific decision. The probability-of-success calc, for example, anchors you to a 10-band base rate (startup survival, marriage longevity, career-pivot traction) and then lets your inputs nudge it; the “answer” is a sanity check against optimism bias, not a prophecy.

Calibration and worked examples

Every calculator on the site ships with a worked example using the displayed defaults. The example lives in the long-form content section below the calculator and walks through the formula step-by-step using the exact numbers the calc would produce on a fresh load. This does two things: it lets you sanity-check the math without trusting our JavaScript, and it pins the calculator to a primary-source number (because the example is calibrated to the underlying authority’s published example, where one exists).

When IRS Pub. 17 includes a worked example for a deduction, our deduction calculator matches it to the cent. When CFPB’s amortization documentation walks through a $200,000 mortgage, our EMI calculator reproduces the same payment schedule. When the WHO BMI fact sheet states that a 70 kg / 175 cm adult has a BMI of 22.9, that is the exact number our default-load shows. If the calculator drifts from the authority’s example, the calculator is wrong — full stop. We anchor to primary-source numbers, not aggregator summaries that may themselves be wrong.

Limitations and disclaimers

CalcBold calculators are educational and decision-support tools, not professional advice. Every calculator carries a domain-appropriate disclaimer: tax calcs aren’t a substitute for a CPA, health calcs aren’t a substitute for a clinician, finance calcs aren’t a substitute for a fiduciary, legal-adjacent calcs (RSU tax-lot decisions, IP valuation) aren’t a substitute for an attorney. When a decision matters, talk to someone licensed. Defaults skew US (currency, tax, units, health bands) unless the calculator is explicitly flagged as UK-, India-, or multi-country. Every calc surfaces its jurisdiction at the top of the long-form section so the limitation is impossible to miss.

Annual review and corrections

We sweep the whole catalog at least once per calendar year — the sweep aligns to whichever authority publishes most recently in that domain (IRS in Q4 for the following tax year, HMRC in Q1, CBDT in Q1, vendor AI pricing every quarter). Refreshes touch three things: the formula constants, the default values shown in placeholders, and the snapshot dates in the page footer. When a sweep changes a calc’s headline result by more than 1% we add a corrections note linking to the change. To report a math bug, a stale source, or a missing edge case, email [email protected]with the calculator URL and the input values that reproduce the issue. We respond within 48 hours and ship confirmed P0 fixes the same day they’re verified.

Versioning

This methodology page was last reviewed end-to-end on 2026-05-02. The editorial team behind the site is documented on /about, including who builds it, how it’s funded, and how to contribute corrections. Per-calculator version stamps live in the footer of each calculator page — they’re what you should cite if you’re quoting a CalcBold number anywhere durable. The methodology stamp on this page tells you the editorial framework is current; the stamp on a calculator page tells you the specific math behind that calculator is current.