--- title: "How this site checks itself" url: "https://thecybersecurityplace.com/method/" section: "Method" date: "2026-09-04" reviewed: "2026-09-04" --- # How this site checks itself Every figure on this site is **computed while the page is built**, from a corpus of **14,951** entries drawn from **969** publications between **2014** and **2021**. None is typed by hand. Seven checks run against every page and the build halts if any one fails. As of the last run, **51 of 51** pass. What the checks do **not** do is decide what is **true**: they verify **provenance and structure**, never **accuracy**. The **four known defects** in the corpus are published below, and one conclusion has been **withdrawn**. ## Key points - **No figure is typed.** Remove the corpus and these pages stop building rather than keeping their numbers. - **The evidence is the observed minimum, not the rule.** The floor is 3,000 words; the thinnest page published holds 3,000. - **The defects are published.** Four of them, each found by measuring this record against itself. - **Checks verify provenance, not truth.** A page can pass all seven and still reach a poor conclusion — which happened once, and stayed documented. ## The seven checks Written material enters at the top and reaches a reader only by passing all seven. A failure anywhere stops the build, and a stopped build publishes nothing. - **build** — refuses a figure it cannot compute: if the record does not hold the data, the build stops and there is no page. - **editorial** — refuses a page that cannot carry itself: too short, vocabulary that repeats, the stock phrases of generated prose, sentences of suspiciously uniform length. - **espacios** — refuses two words fused together by a template defect, invisible while writing and glaring on the page. - **rutas** — refuses a promised address that does not exist, including any the menu links to. - **sitemap** — refuses a mismatch between what the site declares it publishes and what it published. - **gemelos** — refuses a page whose plain-text version says something different from the formatted one. - **enlaces** — refuses a page nothing links to, and any broken internal link. ## What it means that a figure is computed Nobody typed it. When a page says a subject appears a certain number of times, that number is produced while the page is built, by counting the entries that match. A typed number stays correct until it silently stops being correct, and nothing announces the moment. A computed one moves when its source moves. There is a stronger property underneath: remove the corpus and these pages do not fall back to their last known numbers, they fail to build. The supporting data is a build dependency rather than a reference. ## The rule, and the observed minimum Stating a threshold is documentation; anybody can publish a standard nobody applies. Below is each threshold beside the worst value observed across the 51 pages measured. | Rule | Threshold | Observed | | --- | --- | --- | | at least 3,000 words | 3,000 | thinnest page 3,000 | | lexical heterogeneity of 0.32 or better | 0.32 | least varied page 0.320 | | between 30% and 45% of headings phrased as questions | 30–45% | observed range 31.3–42.9% | | sentence-length variation of at least 5 | 5 | flattest page 8.2 | Across those pages the checks cover 159,665 words written for this site. One number here is necessarily out of date. The checks run after the build, so a page cannot publish the result of a check not yet performed on it. The report shown was written on the most recent run and describes the run before this one. ## The record these figures come from 14,951 entries from 969 named publications, running 2014-08-07 to 2021-11-30 — 7 years and 4 months. An entry holds a headline, a date, an attribution, subject tags and, for 6,634 of them, a written summary. It is a record of what the trade press published, which is a different thing from a record of what happened. Two consequences follow. Coverage volume reflects editorial attention rather than incidence. And attribution is to the publication rather than any primary source, so a claim repeated by forty outlets appears forty times without becoming more true. ## Four defects, published 1. **Three months of 2016 are thin** — April, May and June hold 741 entries between them against 171 a month across the rest of that year. The shortfall is roughly -229 entries. 2. **The record does not begin gradually** — August to November 2014 hold 88 entries described at 100%; December alone holds 257 at 43%. Collection ran in two modes and nothing here explains the switch. 3. **Under half of the corpus carries a written summary** — 6,634 of 14,951 entries — 44% — have text beyond a headline. The rest exist as a title, a date and an attribution. 4. **The record always lags the present** — The most recent archived entry is dated 2021-11-30. Filing runs behind publication, so there is always a stretch of recent weeks the corpus cannot answer for. The gap closes as material is added; it never reaches zero. ## Why publish them Because a figure cannot be judged without them, and because a collection that conceals its gaps invites the over-reading that produces bad conclusions. Take the thin months of 2016. Any subject total for that year is depressed by them, and a reader comparing 2016 against 2017 without knowing would conclude that attention rose when the collecting merely resumed. The defect does not make the totals wrong. It makes one particular comparison wrong, which is worse, because the comparison looks sound. There is a commercial argument too. A buyer evaluating this record will find its defects within an afternoon of serious work. Finding them listed is reassuring; finding them unlisted, after being told the material is rigorous, ends the conversation and deserves to. ## Can a verified page still be wrong Yes. The checks verify that a figure matches its source and that a page is structurally capable of carrying an argument. They say nothing about whether the source is complete, whether the sample represents anything, or whether the conclusion follows. It failed here once. A page about 2018 explained a fall in publication volume by pointing at a fall in the number of contributing publications. Both figures were computed and both were right. When the surrounding years were measured the explanation collapsed: the following year recovered its publication count while volume stayed low. The conclusion was withdrawn. That page now establishes the shape of the fall, rules out the two obvious explanations and names no cause. The withdrawal is described on the page rather than quietly patched, since a correction nobody can see is not a correction. ## What happens when a page cannot support a claim The claim comes out. A claim on these pages is a computation over the record; if the record cannot answer it, there is no computation to perform, and the page fails to build rather than rendering an approximation. The failure mode is a broken deployment, which somebody notices immediately, instead of a plausible sentence, which nobody notices at all. ## The plain-text twins Every page written for this site publishes a second version in plain Markdown, at the same address with `.md` appended. A check compares the two and fails the build when they disagree, which stops the accessible version from drifting behind the formatted one. Both are generated from the same computed figures; neither is a transcription of the other. ## The far edge of the record The most recent entry is dated 2021-11-30. Nothing after that date is inside what this corpus can answer. A boundary of this kind is easy to read past, because a record gives no signal at its own edge. The final weeks look like ordinary weeks. A subject that was building when collection stopped appears to have stopped building, and one that had not yet emerged appears never to have existed. Every page that touches the boundary says so. ## Is any of this checkable from outside - **What a reader can check now.** Every figure states the record it came from and the period it covers. Every proportion states its denominator. The observed minimums can be compared against the thresholds. Any page can be read against its plain-text twin. - **What requires the data.** Reproducing a specific count means having the corpus, which is available to researchers and buyers on request. Nothing about the arithmetic is proprietary; it is counting. - **What cannot be checked from outside at all.** Whether the collection was even-handed while it was being made. That is a property of decisions taken years ago, and no amount of measurement recovers it. ## What this is not - **Not a guarantee of accuracy.** Seven checks passing means a page is supported and structurally sound, not that the conclusion is right. - **Not an audit.** Nobody independent has reviewed this. The checks are run by the same build that publishes the pages, which makes them a discipline rather than an assurance. - **Not a claim about the news entries.** Archived items carry whatever text their original publisher supplied, attributed to them. - **Not finished.** Checks get added when a class of error turns out to be possible. The one catching words glued together by a template exists because that defect reached publication. ## Common questions **What does it mean that a figure is computed?** That no number on this site was typed by a person. Each one is calculated while the page is built, from the 14,951 entries described below. Change the corpus and the number changes; remove the corpus and the page does not build at all. **Can a computed figure still be wrong?** Yes, and this is the limit worth understanding. Computation guarantees that a figure matches its source. It guarantees nothing about whether the source is complete, representative or well collected. Every defect listed on this page produces figures that are correct and still misleading if read without them. **What exactly do the checks refuse?** Seven things, each with its own script: an uncomputable figure, a page too thin or too repetitive to be worth publishing, words glued together by a template defect, a promised address that does not exist, a mismatch between what the site declares and what it published, a plain-text twin that disagrees with its page, and any page nothing links to. **Do the checks decide what is true?** No. They check provenance and structure, never accuracy. A page can pass every one of them and still reach a poor conclusion from good data — which has happened here once, and is documented rather than removed. **How current is the verification shown?** It is from the previous run: the checks execute after the build, so a page cannot publish the result of a check that has not yet been run on it. The report shown was written on the most recent run and covers 51 pages. **Why publish the corpus defects?** Because a reader cannot judge a figure without them, and because a collection that hides its gaps invites exactly the over-reading that produces bad conclusions. Every defect here was found by measuring this corpus against itself, not by anybody remembering it. **What happens when a page cannot support a claim?** The claim comes out. There is no mechanism for publishing something the corpus cannot carry, because the figure would have nothing to compute from and the build would stop. **Has anything been retracted?** Once. A conclusion about why publication volume fell in 2018 was withdrawn when the surrounding years were measured and contradicted it. The page now establishes the shape of the fall, rules out the two obvious explanations and names no cause. **What are the plain-text twins?** Every editorial page publishes a Markdown version at the same address with `.md` appended. A check compares them and fails the build if they disagree, so the accessible version cannot quietly drift from the formatted one. **Does this cover the news entries too?** No. The 14,951 archive entries collected so far are attributed to their original publisher and carry whatever text that source supplied. The checks described here apply to the pages written for this site. **Can any of this be verified from outside?** Partly, and that is deliberate. Every figure states the corpus it came from and every page states what it cannot support, so a reader with the same data can reproduce the arithmetic. The corpus itself is available on request. **Why go to this trouble?** Because summarising the news is now free and consequently worth very little, while a claim somebody can check has become scarce. The checks are the only part of this site that would be difficult for anybody else to copy. ## Corrections Errors are corrected on the page where they appeared, with the correction described rather than applied silently. Write to editor@thecybersecurityplace.com with anything that looks wrong — particularly a figure, since a wrong figure means either the record or the computation is wrong.