What gets published,
and what gets refused.

A page here begins with a question someone actually asks — "what will this loan cost me in total", "how much of my income is already committed", "when does PMI come off". If two pages would answer the same question, they should be one page, and if a page cannot name the question it answers, it is being written for a search engine rather than a reader.

1. The question is distinct. It is not a rephrasing of a question another page already answers. Near-duplicate pages split the answer and help nobody.
2. The arithmetic is verified independently. Expected values are worked by hand or taken from a regulator's own published example, never recorded from whatever the code returned first. A test that captures current output proves only that nothing changed since.
3. The working is visible. The formula appears on the page, with every variable defined, alongside a worked example using real numbers and an explicit list of what the tool ignores.
4. Every rule traces to its source. Where a figure comes from law or regulation, the document is cited with the date it was read. Where it is industry convention, it is labelled as convention.

All 15 published calculators meet the third condition in full — formula, plain-language walk-through, worked example, assumptions, exclusions and sources. Between them they state 109 assumptions, 109 exclusions and answer 104 questions directly.

The order of those conditions is deliberate. Accuracy comes before usefulness, usefulness before presentation, and presentation before anything to do with search. A page that is beautifully organised around a wrong number is worse than no page. The formulas and the conventions themselves are set out on how we calculate.

Most financial content is sourced to other financial content. A figure appears in a guide, a second guide cites the first, a third cites the second, and by the fourth the original qualification has been lost — which is how the 28/36 rule came to be quoted as though it were law. The rule here is that a citation must be the primary document, and there are exactly two kinds it can be.

Law or regulator

38

First-party contract

7

Secondary sources

0

The first kind is the rule as published: a federal agency's own guidance, or the authoritative text of the U.S. Code and the Code of Federal Regulations. The second kind is narrower and needs explaining. For a question like "how does this issuer calculate a minimum payment", the cardmember agreement is the primary document — a summary of it is not, and the differences between issuers are the substance of the answer.

Anything that is neither is a secondary source and is not cited. That is why the third figure above is computed rather than stated: if a citation to a blog, a content marketing page or an aggregator is ever added, the number stops being zero on this page at the same moment a test fails.

Every one of the 45 citations carries the date it was read and a note saying what it is relied on for. A bare link proves a page exists; it does not show that anyone read the relevant clause.

Retrieval dates are published because regulations are amended. A date tells you how current our reading is and lets you check whether the text has changed since — which is information a reader can act on, unlike a "last updated" stamp that moves every time a build runs. The full bibliography, with every date, is on how we calculate.

The conventional move here is a photograph, a name and a credential — "reviewed by a CFP®", "written by a panel of in-house analysts", a founder with twenty years in the industry. A reader cannot verify any of it, which is exactly why it is so widely used and so frequently untrue.

So nothing of the kind appears on this site. The publisher stands as the author in every page's structured data, no page passes an author name to the schema builder, and there is no review board, no editorial panel and no fact-checker to name. An unverifiable byline adds nothing a reader can use.

That is a real limitation and it should be read as one. Nobody here is regulated or certified to give financial advice, and nothing on this site is advice. The disclaimer sets out what that means in practice.

What replaces a credential is a standard about the content rather than about its author: every factual claim on this site must trace either to a cited primary source or to a computation with a test behind it. That is a stronger commitment than a byline, because it can be checked line by line — and it is the reason the citation and test counts on this page are computed rather than asserted. The same reasoning, at more length, is on about this site.

A list of prohibitions costs nothing to write, so it is worth almost nothing on its own. The right column is what makes this section meaningful: for each ban, either a test that fails the build when it is broken, or an honest admission that nothing but judgement is holding the line.

An invented author, credential, team or review board

Enforced by a test No page passes an author name to the schema builder, so the organisation is the author in structured data. Asserted on every trust page.

Ratings, reviews or testimonials of any kind

Enforced by a test There is no rating, review or testimonial markup anywhere in the source, and a test greps the tree for the schema types that would carry one.

A statistic typed into markup

Enforced by a test Every figure a page states about the site is read from a registry or from a constant a test reconciles against the files on disk.

Structured data describing something not on the page

Enforced by a test Schema is generated from the same values the page renders, and the FAQ markup is built from the questions actually printed on it.

An internal link to a page that does not exist

Enforced by a test Links to unbuilt pages and unpublished tools are dropped rather than rendered, and every related-tool id must resolve.

Affiliate, sponsored or referral links

Enforced by a test No affiliate, sponsored or referral marker exists anywhere in the source, and a test fails the build if one is added.

A convention presented as a legal requirement

Judgement only Where a threshold comes from statute it is cited to the statute. Where it is industry habit — the 28/36 rule, the 43% figure — it is named as habit at every appearance.

A tax figure that has not been verified for the year it claims

Judgement only One tool here models tax: UK Take-Home Pay Calculator. Every rate is cited to the authority that sets it, with the date it was read, and the tax year is named on the page rather than implied. Where a nation sets its own bands — Scotland for Income Tax — the page says so instead of applying the wrong ones.

A page written because a phrase has search volume

Judgement only A page is published when it answers a question a reader has. Empty categories are not routed at all, so the site cannot grow a shell of thin pages.

The last three carry no test, and saying so is the point. Whether a threshold has been labelled as convention rather than law is a judgement about a sentence, and no grep can make it for us. Marking those honestly is more useful than padding the column with tests that do not really check what they claim to.

One prohibition is worth stating twice because it is the one an ad-supported site is most tempted by: no page here exists to occupy a search result. A hundred pages spun out of keyword variations would be a different site and a worse one, and the growth rule is that an empty category is not a page at all.

A publication standard is only credible if something is currently failing it, so here is the count. Of 32 calculation functions in the engine modules, 17 are reachable from a page and 15 are not. The 15 are finished: each is covered by tests in the same suite that guards the tools already published. What they do not have yet is a page, which is the part this policy is about.

That is the intended order. The maths is built and verified first and the page is written second, because a page is a claim about arithmetic that already works — and the reverse order is how a site ends up publishing a calculator whose formula nobody checked. The withheld figure therefore falls every time a tool ships, which is the direction a backlog should move.

Withheld here means genuinely unreachable rather than merely unlinked. There is no URL that resolves to any of these, which matters more than it sounds: a page that is live but unlinked still gets indexed, and "we just haven't linked it yet" is how half-finished content ends up as a site's first impression in search results. Nothing is sitting in the registries as a draft at the moment; everything declared there is live.

One number deliberately included because it would be embarrassing: 0 of those calculations are covered by no test at all. A function nobody has checked is not being held back out of restraint, it is merely unverified, and counting it as the former would turn this whole section into the kind of flattering statistic it exists to avoid. The figure is computed, not typed, so it cannot be quietly rounded to zero.

Some checks a test cannot perform. Keyboard-only navigation, screen-reader output, print layout and performance on a real network have not yet been checked by hand on these pages, and this sentence stays here until they have been.

Errors are inevitable on a site whose entire proposition is arithmetic, and the useful question is not whether one will happen but what happens next. The order is fixed: the engine is corrected, a test is added that would have caught the error, and then every page that quoted the old figure is updated. Fixing the prose while leaving the engine wrong is how a discrepancy becomes permanent.

Each calculator carries the date its formula was last reviewed and a version number that changes when the formula changes. Those two fields are what make a correction visible rather than silent — a page whose maths changed but whose dateline did not is indistinguishable from one that was right all along.

What this page will not do is promise a response time, a review cycle or a schedule of audits. Those are commitments nothing here can measure, and an unmeasurable promise on a page about honesty is self-defeating. What is promised instead is the order of operations above, and that the version number moves when the formula does.

The most useful thing anyone can send is a reproducible discrepancy — the inputs, the figure we showed and the figure you expected. Send it through the contact page.

One note on sending inputs: figures typed into a calculator stay in your browser, but a link that already carries them does send them in its request line like any URL. If you would rather not include real numbers in a report, a scaled-down version of the same case is usually enough to reproduce a bug. The privacy policy sets out the distinction.