Editorial standards

No figure is typed into an article

This is the rule the rest of the method hangs off. An article on this site may not contain a number as literal text. Every figure is a placeholder, resolved when the site is built by running the computation that produces it against a named dataset.

The consequence is narrower than it might sound, and worth being exact about. If a dataset is corrected or a calculation changes, every sentence resting on it updates automatically — there is no second copy of the number sitting in the prose to go stale, and nobody can adjust a figure to suit an argument without changing the code that produces it. What this does not do is check that the figure was computed correctly in the first place, or that the sentence around it describes what the figure actually shows. A wrong formula produces a placeholder that resolves cleanly and reads as authoritative; a correct figure can still sit inside a sentence that draws the wrong conclusion from it. An independent audit of this site found errors of exactly that kind, corrected in the articles they affected — the mechanism prevents numbers from going stale, not from being wrong to begin with.

A check runs before every build and refuses to publish an article that breaks the no-literal-numbers rule: a number typed into prose, a placeholder with no source recorded, a source recorded for a figure the article no longer uses, or an illustration presented without saying that it is one. That check is automatic and mechanical. It is not a substitute for an independent arithmetic review, a check that a dataset update has not changed what an article's prose should say, or a professional legal or financial certification — none of which this site claims to provide.

Where the numbers come from

Every article carries a record of which computation produced each of its figures and which dataset that computation read. The datasets are built by scripts from published sources — index history, inflation series, government bond yields and total-return indices, policy rates, bank deposit and lending rates — and each is checksummed against the file it was built from.

Where a dataset has a limitation, it is recorded with the dataset and stated on the pages that rest on it. Some of those limitations are substantial: a series that begins later than the article's question would like, a record containing one large shock rather than several full cycles, overlapping windows that are far from independent observations. Those are said on the page rather than buried here.

Testing a claim, and the control

Where an article tests a strategy, it also tests the opposite of that strategy. A rule that appears to work is not evidence until the mirror image of it has been run on the same data — otherwise a result that is really about the market gets reported as a result about the rule.

That has changed conclusions here more than once. The clearest case is an article about holding money back to buy market falls: the strategy looked marginally better than investing steadily, until the exact opposite rule — deploying on market rises — produced the same answer. The finding was that the timing rule does not matter, which is not what the first test appeared to say.

What we refuse to publish

An invented figure, under any circumstances. If a number is needed and no trustworthy source for it exists, the article does without and says what is missing. A confident wrong number is the worst failure this kind of site can have, because it is indistinguishable from a right one.

This applies to statutory particulars as much as to arithmetic. Several articles here decline to reproduce a tax threshold, a regulatory limit or a scheme condition, and send the reader to the issuing authority instead. Those rules change, and a half-remembered provision repeated confidently is worse than an admission that we have not verified it.

It also applies to conclusions. Where the data cannot settle the question, the article says so. Several articles here end without an answer for exactly that reason.

Independence

No sponsored articles, no paid placements, no public-relations copy, no affiliate links to financial products, and no arrangement of any kind with a product issuer. Nobody can buy a mention, a conclusion, or a place on a list.

The site is funded by advertising. Adverts are selected and served by a third party, not by us; an advertiser does not see an article before publication and has no route to influence it. An advert beside an article is not an endorsement of anything, in either direction.

Corrections

A reported error gets checked against the source, not filed. If it is wrong, it is fixed.

Where a correction changes a figure or an argument that a reader may already have acted on, the change is noted on the page rather than made silently — the point of publishing the method is lost if the record of what it produced can be edited invisibly. Where the fix is a typo, it is simply fixed.

Report an error here. The page it is on, and the number you expected instead, are the two most useful things to include.

What has not been checked

Both of those are stated here because a standards page listing only strengths is advertising. The rule this site applies to its own claims is that a claim may only be raised when the work behind it exists — and that rule binds this page hardest of all.