Our process

This page describes how we work. We publish it as a form of accountability rather than as a credential. If you read a piece here that does not appear to meet the standard set out below, we want to hear about it, because a published process is only worth publishing if we are held to it.

There are seven stages. Each one has a gate: a condition that must be satisfied before the piece moves forward. No piece skips a stage because a deadline is close, and no piece skips a stage because an editor is fond of the topic. The process exists precisely to resist those two pressures, which are the ones that actually occur.

Not every important event becomes a piece here. Three criteria apply at commissioning: public interest, originality, and what we call the contribution test.

Public interest means the story matters to people not directly involved in it, because it has consequences, illuminates a pattern, or changes how something is understood. Originality means the piece adds something that does not already exist: new reporting, a new synthesis, or an angle that changes what existing coverage implies. The contribution test asks one question. If we do not publish this, what does a reader lose? Where the honest answer is "not much", we do not publish it.

Timeliness alone passes none of these. A story that is urgent but derivative, repackaging what wire services already carried, does not meet the bar. We would rather publish two weeks later with better sourcing and a better argument than produce a quick take that adds nothing to the public record.

Research begins with a hierarchy. Primary sources sit at the top: official documents, legislative records, court filings, datasets, peer-reviewed studies, and direct testimony from people with direct knowledge. Secondary sources, meaning reporting by other outlets, academic commentary and published analyses, are useful context and do not substitute for primary material when a material claim rests on it.

Expert sources are consulted for interpretation and for identifying what is and is not established in a field. We name experts wherever possible and give their affiliation and relevant expertise. We do not write that "experts say" without saying which. Where a source asks for anonymity we require a credible reason, meaning a real professional or personal risk, and we tell readers the nature of that reason without identifying the person.

No single source is sufficient for a significant factual claim. Where a claim can be supported by only one source we say so in the piece. Where it cannot be independently verified at all, we say that too.

Fact-checking is not a single pass at the end. It runs through research and drafting, and it includes a dedicated review before any piece is cleared.

What gets checked: every quantitative claim, including statistics, dates, figures and rankings; every attributed quote, against a transcript, recording or contemporaneous notes; every reference to a named individual, including their title and role at the time of publication; and every causal claim, where the logic of the argument is read against the evidence offered for it.

Where a claim cannot be verified the policy is fixed. We do not publish an unverified claim as though it were established. We hold it until verification is possible, frame it explicitly as unverified, or remove it. Publishing under uncertainty without disclosing the uncertainty is a way of misleading readers even when the claim later turns out to be true.

A draft enters editorial review when the writer considers it complete and the research behind it is documented. Review is layered, and the order is deliberate: structure first, language second.

Structural editing addresses the argument. Is the central claim clear? Does the evidence support it, or is there a gap between what the writer asserts and what the sources establish? Is this the right shape for this argument, or does the order of sections lead a reader somewhere the evidence does not go? This is where most of the work happens, and three structural revisions before line editing begins is not unusual.

Line editing addresses language: precision, clarity and concision. We edit for accuracy of expression, asking whether a sentence says exactly what the writer means, and for readability. We do not edit for style alone, and we do not impose a house voice over the writer’s own. A good line edit makes the writer clearer. It does not replace them.

Before clearance a final reviewer who has not been involved in drafting or editing reads the piece against a fixed checklist. Is every factual claim traceable to the file? Does the headline represent the content? Is any language stronger than the evidence warrants? Has every identifiable individual had an opportunity to respond to material claims about them?

That last one is taken seriously. Where we are about to publish a material claim about a named person or organisation we contact them before publication and give a reasonable deadline. We include their response, or we note that they declined or did not reply. We do not give any organisation editorial approval over what we write. A right of reply and a right of control are different things, and conflating them is how access journalism starts.

We publish when a piece is ready rather than when a slot opens. We do not hold a finished piece indefinitely, and we do not publish an unfinished one to meet a schedule. Where a story is time-sensitive and the full piece is not ready, the honest response is a shorter, clearly scoped piece we can stand behind.

Headlines are held to the standard of the body. We do not write a misleading headline and qualify it in the text, and we do not use framing that implies more certainty than the reporting supports. A headline that earns attention by overstating the story is not a success. It is a small editorial failure multiplied by everyone who reads it.

We respect an embargo that serves a legitimate purpose, such as coordinating a scientific announcement, giving a government time to respond to a data release, or enabling simultaneous coverage of a shared event. We do not respect an embargo used to manage a narrative or to delay accountability reporting.

Errors are handled at three levels, and the level decides how visible the correction is.

We do not quietly delete passages, change claims without disclosure, or update an article with new information without saying that we have. Where a piece is updated because the story moved on rather than because we were wrong, we note the update date and summarise what was added, so the distinction between a correction and a development is visible rather than inferred.

If you believe we have made an error, or you hold information that contradicts something we have published, tell us. Every such message is read by an editor.