Research methodology

Research starts from the reader’s question in the reader’s words — “what does a prosecutor actually decide?”, not “prosecutorial discretion”. Before searching, we write down the question, the jurisdictional scope, what would count as an answer, and what would falsify a draft answer.

Comparative work fails when one country’s arrangement becomes the mental template. We counter that by identifying the function first, then finding at least two systems that perform it through structurally different arrangements, and only then describing how any individual system does it.

This is why the jurisdictional-variation section of a guide is drafted early rather than appended at the end.

1. The legal instrument that creates the institution or the power — read, not summarised.
2. The institution’s own official material, treated as self-description and attributed as such.
3. Official oversight material — inspectorate reports, ombudsman findings, audit reports, parliamentary evidence. Frequently the best structural descriptions available, because their job is to describe how something actually works.
4. International-organisation comparative material.
5. Academic comparative literature, for analytical vocabulary and the known limits of comparison.
6. Archives and museum collections, for historical claims.
7. Journalism, last, for contemporary context only, always attributed.

If research begins at step seven and works backwards, the framing of the reporting survives into the page. We start at step one.

A source is recorded only after it has been read at the relevant point, and recording includes writing down what it establishes and what it does not. If that note cannot be written specifically, the source has not been read carefully enough to cite.

The URL is verified at the moment of recording and the date is stored. Constructing a plausible URL from a pattern is prohibited — see research and sources.

Different jurisdictions

Not a conflict at all. Both claims are scoped and the variation is presented.

Primary source vs secondary summary

The primary source governs, and the discrepancy is worth noting if the summary is widely repeated.

Law as written vs inspection findings on practice

Both are presented, with an explicit statement that formal rule and observed practice differ. That gap is often the most instructive thing on the page.

Genuine scholarly disagreement

Positions are presented with attribution, the paragraph is marked as disputed, and we do not adjudicate.

We never resolve a conflict by picking the more convenient source or splitting the difference.

Failed research is recorded rather than discarded. Anything we could not establish goes into the page’s uncertainty list and is shown to you — a structure that changed on a date we could not source, an institution whose current mandate could not be confirmed, a widely repeated figure whose original source could not be located.

Publishing the gap is better than closing it with an assumption, and it tells a future editor exactly where to work.

Fact-checking is a separate pass from editorial review, performed against the source records rather than the author’s memory of them. Every sentence containing a checkable claim is matched to a source; each source is reopened and confirmed to support the claim at the level of specificity stated; numbers, dates, names, and institutional titles are checked character by character. Anything that fails is removed, rescoped, or moved to the uncertainty list.

Safety review follows, for law enforcement, investigations, forensics, and public safety. It asks a different question: not whether the content is true, but what it could be used for.

• Summarising other explainer sites. If the only support for a claim is that other sites say it, the claim is unsourced.
• Reasoning from what seems institutionally sensible. Systems are frequently organised in ways that are not the obvious design.
• Filling a structural gap because a section looks thin. A thin section with an honest scope note is correct; an invented one is not.
• Generating plausible detail. Plausibility is the failure mode, not the standard.