This page sets out the standard our published work is held to. It exists so that a reader can tell the difference between something we have confirmed, something we judge to be probable, and something we are merely reporting that others have said.
Source grading
Every source behind a published claim is graded twice and independently: once for the reliability of the source itself, and once for the credibility of the specific piece of information. A source that is usually dependable can still pass on something doubtful, and the two ratings must not be collapsed into one.
Reliability of the source
| Grade | Meaning |
|---|---|
A | Completely reliable. No doubt of authenticity, trustworthiness, or competence; a history of complete reliability. |
B | Usually reliable. Minor doubt; a history of generally valid information. |
C | Fairly reliable. Doubt exists; has supplied valid information in the past. |
D | Not usually reliable. Significant doubt; has supplied valid information at times. |
E | Unreliable. A history of invalid information. |
F | Cannot be judged. No basis on which to assess reliability. |
Credibility of the information
| Grade | Meaning |
|---|---|
1 | Confirmed. Corroborated independently by other sources; logical; consistent with other information on the subject. |
2 | Probably true. Not confirmed, but logical and consistent with other information. |
3 | Possibly true. Reasonably logical; agrees with some other information. |
4 | Doubtful. Not confirmed; possible but not logical; no other information on the subject. |
5 | Improbable. Contradicted by other information; not logical. |
6 | Cannot be judged. No basis on which to assess. |
A claim carried as B2 is therefore drawn from a usually reliable source and is probably true but unconfirmed — a very different thing from A1, and it should never be presented to a reader as though it were.
How we phrase probability
Words like "likely" mean different things to different readers, so we fix them to ranges and use them consistently. These terms describe how probable we judge something to be. They are not a statement of how confident we are in the judgement itself, which is a separate axis covered below.
| Term | Approximate range |
|---|---|
| Almost certain | 95–99% |
| Very likely | 80–95% |
| Likely | 55–80% |
| Roughly even chance | 45–55% |
| Unlikely | 20–45% |
| Very unlikely | 5–20% |
| Almost no chance | 1–5% |
We avoid hedges that carry no information, such as "it is possible that" or "some reports suggest," where the intended meaning is one of the terms above.
Analytic confidence
Confidence describes the strength of the basis for a judgement, not how likely the judgement is to be true. A high-probability estimate can rest on a weak evidentiary base, and readers are entitled to know when that is the case.
- High confidence. Multiple independent, well-graded sources; consistent with the wider body of evidence; conclusions depend on few assumptions.
- Moderate confidence. Credible sourcing that is partially corroborated, or sound reasoning that depends on assumptions we have stated explicitly.
- Low confidence. Sparse, fragmentary, or uncorroborated sourcing; plausible alternative explanations remain open. Published only where the subject warrants it and always labelled as such.
Labels used in our reporting
Our compliance statement commits us to distinguishing verified facts from estimates, allegations, and analytical judgements. In practice that means every substantive claim carries one of four labels.
- Verified. Established by primary documentation or by two or more independent sources we have examined directly.
- Assessed. Our analytic judgement, carrying a probability term and a confidence level, with the reasoning shown.
- Reported. A claim made by an identified third party that we have not independently confirmed. The attribution travels with the claim.
- Unresolved. Evidence is contested or insufficient. We say so rather than choosing a side by omission.
Verification practice
- Primary documents are preferred over reporting about them. Where we cite coverage, we link the underlying record wherever one exists.
- A claim is not marked Verified on the strength of multiple outlets repeating a single origin. Independence means independent origin, not independent publication.
- Imagery is geolocated against known terrain and, where possible, chronolocated using shadows, weather records, or construction state. Metadata alone is treated as an unreliable indicator because it is trivially edited.
- Source material is archived at the time of access so that a later deletion or edit does not leave a claim unsupported.
- Machine translation is flagged where it materially affects the reading of a source.
What we do not do
The limits below are binding regardless of what a story would gain from crossing them, and are set out in full in our legal and ethical compliance statement.
- No hacking, unauthorised system access, or circumvention of access controls or security measures.
- No impersonation, pretexting, or covert surveillance of individuals.
- No use of material we know or suspect to have been obtained unlawfully.
- No publication of private personal information where there is no clear public interest in it.
When we get it wrong
Open-source material can be incomplete, outdated, inaccurate, or deliberately misleading, and no methodology removes that risk entirely. Errors we identify or that are reported to us are corrected on the record rather than quietly amended. Every correction is logged, dated, and described in the corrections log.