How this material is produced

Why the protocol is public

A reference site asks readers to rely on it, and reliance without visible method is just reputation. Publishing the procedure lets a reader test whether it was followed, and it commits this site to a standard that can be checked against any individual page rather than asserted in general terms.

The protocol is also the site's own subject matter. Much of the material here concerns documenting method, and a resource that recommended documentation while concealing its own would not be worth reading. Where the procedure has been departed from on a page, that page says so.

Scoping a page before writing it

Every page begins with a written scope: the object being described, the population or setting it applies to, the period covered, and the questions the page will not attempt to answer. The scope is fixed before drafting because a page written first and scoped afterwards expands to whatever material was easy to find.

The scope also determines whether a page should exist. Where the available documentation is too thin to describe a procedure without guessing, the page is not written. A gap that is honest is more useful to a reader than a page assembled from inference, and the fields not yet covered are listed on the project page rather than hidden.

Source hierarchy

Sources are admitted in a fixed order of preference, and the order is applied rather than nominal. First, primary documents: legislation and statutory instruments, official guidance, statistical methodology notes, register filings, court and regulator decisions, and the published documentation of a dataset. Second, peer-reviewed research and official statistics, read for their own methodology sections rather than their abstracts. Third, professional standards and codes issued by recognised bodies. Fourth, established reporting by identifiable outlets, used only where it records a fact that can be traced to one of the layers above.

Material that does not qualify as a source includes vendor marketing, undated web pages, aggregator summaries of documents that are themselves available, and social posts, except where a post is itself the object being described. Where a claim rests only on such material, it is either dropped or stated as unverified.

Every source used is recorded with its title, publisher, publication or access date and, where the document is stable, a reference precise enough to locate the passage. Access dates matter because registers and portals change in place: a figure quoted here reflects the source on the day it was consulted, not necessarily today.

How claims are checked

Factual claims are traced to a specific source before publication, and numerical claims are re-derived from the original rather than copied from a secondary summary. Where a figure is quoted, the unit, period, population and definition are checked to make sure the sentence around it does not extend the figure beyond what it measures.

Descriptions of legal duties receive additional scrutiny, because they are the easiest material to get subtly wrong and the most consequential when that happens. Such passages are checked against the instrument itself, stated at a level of generality that does not amount to advice, and accompanied by the standing note that a specific situation needs a qualified professional.

Statements about what tools or techniques can achieve are written to the weaker end of the evidence. Where a capability is demonstrated only in controlled conditions, the page says so; where practitioners disagree, the disagreement is described rather than resolved.

Separating record, inference and open question

Three registers are kept distinct throughout the site. A record is something a document establishes. An inference is a reading of records that is reasonable but not established. An open question is something the available sources do not settle.

The distinction is carried by the language rather than by a label. Records are stated plainly; inferences are marked by construction — the records are consistent with, this suggests, a plausible reading is; open questions are collected explicitly at the end of a page. A reader should never have to guess which register a sentence belongs to, and where that is unclear it is a defect to be reported.

Numbers, tools and arithmetic

Figures on this site are illustrative unless a source is named beside them. Where a page contains an arithmetic tool, the formula is shown, the inputs are described, and the output is labelled as an arithmetic illustration rather than an assessment of any real survey or dataset.

The tools run entirely in the reader's browser. Nothing entered into them is transmitted, stored or logged anywhere, because the site has no server-side processing, no analytics and no third-party scripts of any kind.

Examples, and the rule against invented facts

Illustrative material is used throughout the site — a count of local authorities, a sample size, a percentage change — because a procedure is easier to follow when it has something to operate on. The rule governing it is absolute: an illustrative figure is never dressed as a finding. Where a number has a source, the source is named beside it with its date; where it is there to make arithmetic visible, the surrounding sentence says that it is an example.

Two things are therefore never written here. The first is a statistic that looks verifiable and is not: a fabricated percentage attached to a real institution would be indistinguishable from a real one to a reader, and would damage precisely the practice the page is describing. The second is a named individual. Where a role has to be described, it is described as a role — a verification editor, a data publisher, an official who signed a memo — and no invented person, quotation or case study appears anywhere on the site.

Naming institutions without implying endorsement

Bodies are named when naming them is the most precise way to describe something: a catalogue, a statistical series, a register, a supervisory authority, a legal framework. The naming is descriptive and neutral, and it carries no claim of membership, accreditation, supervision, partnership or approval in either direction. Where a body's practice is described, the description is taken from that body's own published documentation rather than from commentary about it.

Institutions are also not used as an argument. That a statistical office publishes a series says nothing about whether the series answers the question a reporter is asking, and a page that leans on the publisher's reputation instead of its methodology note has skipped the step this site exists to describe. Where a body's own documentation is unclear or silent, the page says so rather than filling the gap.

External sources and the absence of links

No page on this site links out, and nothing is loaded from another domain. That is partly a privacy decision: a reader should be able to consult a reference without a request going anywhere else. It is also a maintenance decision, because an archive of external links decays silently, and a dead link in a reference page is worse than no link at all — it suggests a source was checked when it can no longer be seen.

The alternative is precision in the text. An instrument is named so that it can be found by name; a series is identified by its publisher and title; a register is described by what it holds. Where a source has moved or been withdrawn since a page was written, the review cycle is what catches it, and the access date recorded against every source is what makes the change visible rather than invisible.

Register and sentence construction

Language is treated as part of the method rather than as decoration on top of it. Verbs are chosen to match the strength of the evidence: rose alongside where two series moved together, is consistent with where a mechanism is plausible and untested, caused only where a design supports it. Quantities are given with their unit, period and population, and comparative claims name what they are being compared against.

Three habits are avoided deliberately. Intensifiers that add no information — significant, dramatic, alarming — are cut, because they invite the reader to supply a magnitude the evidence does not contain. Generalising from a stated sample to a population is treated as a factual error, not a stylistic choice. And the passive construction that hides who did something is rewritten, since in descriptions of procedure the identity of the actor is usually the point.

How reader reports are triaged

Reports are read in the order they arrive and acted on in order of consequence. A report that a figure or a legal description is wrong is checked first, because a reader could act on it today. A report about ambiguity is next, then a report about a broken internal link, then a suggestion about scope, which is logged against the relevant page rather than answered immediately.

Every report produces the same first step regardless of category: the claim is traced back to the document it was written from. That check is recorded even when it confirms the existing text, so that a second report about the same sentence does not repeat work already done. The classification and the outcome of each report are kept with the page, which is how a pattern — a section that is repeatedly misread — becomes visible and gets rewritten.

Accessibility inside the editorial procedure

Accessibility is handled as an editorial step rather than as a technical afterthought, because most of what makes a page usable is written rather than coded. A heading order that does not skip levels, link text that means something out of context, a table with a caption and row headers, a text alternative that states the finding of a diagram instead of describing its decoration: each of those is decided while the page is being written.

Two consequences follow for content. Any point made only in a graphic is also made in the surrounding prose, so nothing depends on seeing an image. And any figure shown in a table is stated in the text where it carries the argument, so a reader navigating a dense table by screen reader is not the only route to the finding.

Review cycle

Each page carries the date of its last review and is scheduled for rereading at least annually. A review is triggered earlier by a change in legislation or official guidance, a redefinition of a statistical series, the relocation or withdrawal of a public register, the publication of research that changes the state of knowledge, or a reader report.

A review has three possible outcomes: confirmation, in which case the review date advances; revision, in which case the change is noted at the foot of the page; or withdrawal, where a page can no longer be maintained accurately, in which case it is removed and the removal is recorded on the sitemap page rather than leaving a silent gap.

Corrections

A factual error is corrected explicitly. The corrected page carries a dated note stating what was wrong, what is now correct and whether the change affects the surrounding conclusions. Earlier text is not replaced silently, because a reader who acted on the previous version is entitled to see that it changed.

Reports should name the page, quote the sentence in question and give a source that can be checked. Reports are acknowledged within five working days; where a report is not accepted, the reason is given. The distinction between a correction and a difference of interpretation is applied honestly: an interpretation this site declines to change is recorded as a stated disagreement rather than dismissed.

What this protocol cannot deliver

A published procedure reduces error; it does not eliminate it. Documents are misread, sources are superseded between review dates, and general descriptions of duties will not fit every situation. The protocol is a commitment about process, not a warranty about content.

Two limits follow directly and are repeated on every page. This site is not accredited by any authority and issues no qualification. And its material is general information rather than personalised legal, professional or data-protection advice: for a specific case, the applicable rules in force and a qualified professional are required.