Method guides

How to read these guides

The guides are built for selective reading. The section index stays visible so that a specific step can be reached without working through the whole text, and the sections are numbered so that a reference remains stable when the page is later revised.

Each guide separates three levels. The first is procedure: what to do, in what order. The second is judgement: the decisions inside the procedure that cannot be reduced to a rule, and what should be recorded when they are made. The third is limits: what the method cannot establish however carefully it is followed. Collapsing these levels is the most common failure in writing about investigative technique, because it turns a set of trade-offs into a recipe.

Why four guides and not forty

The archive is deliberately small. Reference material earns its value by being maintained, and maintenance means rereading a page whenever legislation changes, a register moves, a statistical series is redefined or relevant research appears. An archive that grows faster than the capacity to revise it produces pages that were accurate on the day of publication and misleading a year later.

These four cover the sequences that most often go wrong in practice: taking a dataset from download to publication, interrogating a published figure, verifying material found in public, and keeping the records that make an investigation defensible. Further guides will be added only where there is enough documentation to describe a procedure without guessing.

Which guide answers which question

The four titles overlap deliberately, because the work does. If you are unsure where to start, the question in the middle column is the one the guide was written to answer.

Guide	The question it answers	What it will not do
From open data to a published story	I have a dataset. What is the order of operations from here to publication?	Choose a tool or a programming language for you
Reading statistics without being misled	Someone has published a figure. What do I need to know before quoting it?	Assess any real survey or produce an estimate
The verification chain	Material has appeared in public. In what order do I test it?	Describe any means of reaching material that is not public
Documenting an investigation	What records must exist for a finding to survive being challenged?	Advise on a legal risk in a specific piece of work

What the guides assume, and what they leave out

They assume an adult reader who can open a spreadsheet and follow an argument about numbers, and nothing beyond that. Where a statistical idea is needed it is defined in place; where a term recurs across the site it also appears in the terminology list. No guide requires a particular application, service or programming language, and none is written around one, because a method that only works inside one tool is a habit rather than a method.

Three things are absent by design. There are no step-by-step instructions for named software, which date faster than anything else and would turn a reference into a manual. There is no jurisdiction-by-jurisdiction account of access or media law, which would be advice this project is not in a position to give. And there is no worked example built on invented figures: where a number appears without a named source, the surrounding text says that it is illustrative.

One arithmetic tool

The guide on reading statistics carries a margin-of-error calculator, and the guide on open data links to it. It performs a single standard calculation for a proportion and exists to make one point concrete: precision degrades quickly in subgroups, and two estimates that differ by less than their combined intervals do not establish a difference.

It is an arithmetic illustration only. It does not assess a specific survey, does not account for non-response or question wording, and produces no advice.

Revisions and corrections

Every guide shows the date of its last review. A substantive update is noted at the foot of the page with the nature of the change; a factual error is corrected explicitly rather than replacing the earlier text silently. Reports of errors are welcome through the published contact details.