| Check | What it runs | Required result |
|---|---|---|
| Worked-example run | All 663 calculators at their prefilled inputs | Every output finite and defined |
| Boundary & stress | 663 × 6 extreme values, plus zero | No infinity, no undefined, no stall |
| Independent recompute | Core formulas via a second, different method | Both methods agree |
| Link integrity | Every internal link and schema URL | All return 200 |
1. Define the financial question behind each calculation
We specify the cash-flow timing, units, rate convention, compounding period, inclusions and exclusions before coding. APR, interest rate, APY, return and yield are not interchangeable.
2. Prefer primary evidence
Current rules and definitions are checked against regulators and public agencies such as the IRS, CFPB, SEC, FDIC, Federal Student Aid, SBA and BLS. A source supports a definition. It does not make an estimate personal advice.
3. Test the arithmetic
Every formula is run against representative inputs, zero boundaries, decimals and large values. Division and logarithm boundaries resolve safely. Display rounding never changes the underlying calculation. The automated suite exercises all 663 calculators at their worked-example inputs, at zero, and at 6 extreme values each, 3978 calculator and input combinations, and fails if any result returns infinity, if a result that should be defined becomes undefined, or if any calculator takes long enough to stall a browser.
4. Recompute independently
Core formulas are checked against a second implementation that reaches the answer a different way: closed-form loan payments against a month-by-month simulation, bond price against explicitly discounted cash flows, compound growth against a period-by-period accumulation, and federal tax against an independent bracket walk. An error in the formula and an identical error in the check would both have to occur for a wrong figure to pass.
5. Separate estimate from interpretation
The result is arithmetic from the values entered. The page explains what the number can answer, what it cannot answer and which assumption deserves a second scenario.
6. Date and version current rules
Tax brackets and contribution limits are labeled by tax year. Material formula or rule changes receive a new review date and are propagated to dependent pages.
7. Protect user inputs
Calculations run in the browser. Share links include numeric inputs only. Do not enter account numbers, Social Security numbers, credentials or other sensitive personal information.
8. Version formulas, not just pages
Each calculator carries a formula version alongside its review date. A change to arithmetic, rounding, a rate convention or a bracket table increments the version and triggers a re-check of every page that depends on it. A change to wording does not. Separating the two means the review date signals what a reader needs it to signal: whether the calculation itself has moved.
9. Publish the worked example, not a screenshot
Each page shows a result computed at prefilled inputs, and every input is named beside it. That makes the example reproducible: anyone can re-enter the same values and obtain the same figure, or dispute it. A screenshot of a result proves nothing and cannot be checked, which is why none appear here.
10. Show sensitivity , not a single number
A single result conceals how fragile it is. Each calculator publishes the headline figure recomputed across a range of the input that drives it, and where two inputs are both meaningful, across a grid of both. This is what turns an answer into a decision aid, because the useful question is usually not what the number is but how much it moves.
11. State the boundary of every estimate
Each calculator carries an interpretation boundary describing the conditions under which the model stops being a reasonable approximation: fee structures it does not capture, rules that vary by jurisdiction, timing conventions it simplifies. An estimate without a stated boundary invites use in situations it was never suitable for.
12. Separate what is computed from what is cited
Figures produced by a formula are labelled as computed and are reproducible from the inputs shown. Definitions, rules and current-year figures are attributed to a primary source with a link. The two are never blended, because a computed figure carries the authority of arithmetic and a cited figure carries the authority of its source, and conflating them borrows credibility that has not been earned.
Frequently asked questions
How is a calculator built and published?
Twelve steps take a financial question from definition to a published, tested calculator, starting with fixing the cash-flow timing, units and rate convention before any code is written.
What checks does every calculator pass before publication?
A worked-example run of all 663 calculators, 663 × 6 extreme values plus zero, an independent recompute of core formulas by a second method, and a link-integrity check where every internal link and schema URL returns 200.
How are formulas checked for errors?
Core formulas are recomputed by a second implementation that reaches the answer a different way, for example closed-form loan payments against a month-by-month simulation. An error would have to appear in both methods for a wrong figure to pass.
Why are APR, interest rate, APY, return and yield handled separately?
Because they are not interchangeable. The rate convention, compounding period, inclusions and exclusions are specified before coding so each term means one thing.
Which sources are treated as primary evidence?
Current rules and definitions are checked against regulators and public agencies such as the IRS, CFPB, SEC, FDIC, Federal Student Aid, SBA and BLS. A source supports a definition, but no source turns an estimate into personal advice.
What does a formula version bump mean?
A version bump means the arithmetic itself changed. Formula versions move independently of wording changes, so rewording a page never bumps one.
How are rules that change by tax year handled?
Each rule that changes by tax year carries its own review date, which is propagated to every dependent page instead of being patched on a single page.
Why does every headline result show a range?
Because a single number conceals how fragile it is. Every headline result publishes its sensitivity across a range of the input that actually drives it.