How we work

Every figure is dated. Every claim is attributed to its source. This page explains how we check what we publish, what our labels mean, and how to tell us when we get something wrong.

Our sourcing standard

A key claim, and every number, needs one primary source or two independent reputable sources. Two outlets repeating one wire story count as one source.

Primary sources come first: official gazettes and regulators, ministries and official media offices, company announcements about what the company itself did, peer-reviewed research and official statistics.
Reputable reporting comes next: international and regional outlets with editorial standards, such as Reuters, AP, the BBC, The National, Khaleej Times, Gulf News and Arab News, and national news agencies.
We never cite social media posts, unattributed reports or AI-generated summaries. They are leads to check, nothing more.

Every figure carries a date. We never invent a date or present an estimate as a fact. If a source gives only a month, we show only the month. A news page is dated to the day the story happened, taken from its sources.

We keep what has been announced separate from what is operating. We attribute superlatives such as "world's first" to whoever made them. Estimates are labelled as estimates, with who made them.

Verification codes

Facts in our fact base and tracker carry a code that says how the source was checked.

[F]: Fetched. An editor opened and read the source page, and the fact matches its text.
[S]: Search-verified. The source's text was confirmed through a search index or a reliable summary, but the page itself could not be opened (for example, it was blocked). We upgrade it to [F] at the next review where we can.
[2nd]: Secondary. The fact comes from a reputable outlet summarising a primary report. We confirm it against the primary source.
TO VERIFY: We saw the claim but could not confirm it to our standard. We do not publish it as fact.

Labels for forward-looking claims

Content about the future carries a visible label, so you know what kind of statement you are reading:

Verified fact: A measurement or event, confirmed by a reliable, dated source.
Official target: A goal announced by a government or official body, with a date or an amount.
Company claim: What a company says about itself, its products or its projects.
Expert view: A forecast or assessment by an expert or a specialist body. It is an opinion, not a measurement.
Speculative: A statement about the future that cannot be measured today, such as estimates of the chance of catastrophe.

Tracker statuses

Statuses describe evidence, not people. Each one carries an "as of" date. When the evidence would support two statuses, we choose the more cautious one. Achieved and Stalled need one primary source or two independent reputable ones.

Achieved: The specific outcome claimed has happened, within the deadline if there was one.
On track: The deadline has not passed, and recent evidence shows a specific, checkable milestone or a measured trend holding at the claimed rate. A bare statement that something is "on track" is not enough.
Emerging: Early, credible signs that the outcome is beginning, such as first deployments or pilots, but too little to judge pace or scale.
Not yet: The outcome has not happened, the deadline has not passed, and there is no substantial evidence either way. This is the default for long-range forecasts.
Stalled: The deadline has not passed, but progress has stopped or reversed: for example a reported delay, suspension or withdrawal of funding, or no visible progress for 12 months.
Missed: The deadline and a grace period have passed without evidence that the outcome happened, or the claimant has said it was not met. Missed is not failure: we always say what did happen.
Superseded: The claimant has formally replaced the claim with a new one, or withdrawn it. The entry is frozen and links to its successor.
Reference: A label, not a status. Definitions, measurements, expert estimates and attributed superlatives cannot be "achieved", but we review them for accuracy and currency.

Modifiers

Disputed: Credible sources disagree about whether the claim has been met, or how to read it. We show both readings and their sources.
Company-reported: The latest evidence comes only from the claimant, or from a party with an interest in the outcome.
Evidence dated: The newest evidence is more than 180 days old, so the entry gets priority review.

See them in use in the tracker.

How pages are reviewed and refreshed

Every draft goes through an automatic check for filler and hype, then a human edit.
Only an editor with publishing rights can approve a page. Assistants, including AI tools, can draft but cannot approve or publish.
Pages show a "last reviewed" date.
Tracker entries are reviewed each month when they fall due, all of them every quarter, and again after major annual reports. A deadline passing or a major official announcement triggers an earlier review.

The daily fact-check

Every day, an automated check looks over what we have published. It tests whether cited links still work, whether any date described as past is actually in the future, and whether the English and Arabic versions give the same figures. It also flags statements that something is "not available" or "not confirmed", so we research them again.

A second, AI-assisted step can run on a random sample of claims, weighted towards figures, negative statements and sensitive topics. One AI model searches the web, preferring primary and local Gulf sources, to try to show that a claim is wrong or out of date. Every quote it finds is checked against the real page. A second model, from a different developer and without web access, then weighs that evidence.

The findings go to our editors. The fact-check never changes a published page by itself. An editor decides what to correct.

Arabic and English pairs

Each article is two linked documents, one in each language, and both go live together. If one language is missing, the page is not published.

Arabic is written for Arabic readers in Modern Standard Arabic, not machine-translated word for word. Figures and sources must match the English exactly. When the English changes, the Arabic is flagged for an update.

Review by a native Arabic speaker is part of our process. That team is still being set up, so some Arabic wording may be refined.

How we use AI

We use AI tools as assistants, and we say so:

They help with research and first drafts.
Our news monitor reads public news feeds and files drafts only. An AI step may rewrite a feed excerpt in calmer language, with instructions not to add names, dates or numbers.
They may suggest tags for people, organisations and places. An editor accepts or rejects each one.
They help with the daily fact-check described above.

AI tools cannot publish anything. Every page goes live by an editor's decision.

Corrections and right of reply

When we get something wrong, we correct it promptly and say so. Corrections carry a date. We do not quietly delete the history.

Every tracker change is logged with its date and reason. Accepted corrections are marked "Correction".

A public channel for corrections is being set up and will be listed on our Contact page. When you write, tell us the page, what you think is wrong, and a source if you have one.

Right of reply

Before we mark a named organisation's target as Stalled or Missed, we look for its latest public statement and record it. If an organisation sends us a statement or evidence, we review it and publish a factual summary of its position alongside ours. We do not remove accurate, sourced entries on request.