What Counts As Evidence
We distinguish between primary sources, secondary reporting, interpretation and speculation.
Primary material can include:
- official records;
- archival documents;
- correspondence;
- datasets;
- photographs;
- contemporaneous reports;
- technical records;
- first-hand testimony; and
- other material created close to the event or issue being examined.
Secondary sources can provide useful context, but they do not automatically carry the same evidential weight as the underlying record.
For every important source, we ask questions such as:
- Who created it?
- When was it created?
- For what audience?
- What access did the source have to the facts?
- Is it contemporary with the event?
- Is it complete?
- Has it been copied, edited or summarised?
- Does another source support or contradict it?
- Is there any reason to treat its account cautiously?
A source can be genuine and still be mistaken, incomplete or misleading.
How An Investigation Moves
1. Define the question
We begin with a specific claim, contradiction, sequence or evidential gap that can realistically be tested.
The investigation should not begin by assuming that a particular person, institution or mechanism caused the problem.
2. Build the research record
We collect the strongest available material relevant to the question.
Where possible, we prefer original documents, official records, primary datasets and contemporaneous material over later summaries.
Secondary reporting may be used where it adds context or helps identify the underlying record.
3. Check provenance and context
Documents are assessed in context.
Depending on the material, this can include checking:
- provenance;
- dates;
- signatures;
- stamps;
- catalogue references;
- amendments;
- redactions;
- publication history;
- metadata; and
- whether the document is an original, copy, transcript or later reproduction.
Where authentication cannot be established, that limitation is retained.
4. Build the chronology
Dated material is placed into sequence so that the order of events is clear.
This helps distinguish between:
- what was known at a particular time;
- what happened later;
- what was reported retrospectively; and
- what may only appear connected because of chronology.
Sequence alone is not treated as proof of causation.
5. Compare competing accounts
Where sources disagree, we preserve the disagreement rather than forcing them into a single version.
We ask:
- whether one source had better access;
- whether one account is later than another;
- whether terminology changed;
- whether records are incomplete;
- whether the disagreement can be resolved; and
- what evidence would be needed to resolve it.
If more than one explanation remains compatible with the evidence, the investigation should say so.
6. Test claims against sources
Before drafting, significant claims are checked against the sources that are supposed to support them.
We use structured research notes and claim-source controls to distinguish between:
- confirmed facts;
- disputed claims;
- unsupported statements;
- interpretation;
- working hypotheses; and
- unresolved questions.
The aim is to prevent an inference or research hypothesis from quietly becoming a published fact.
7. Draft without pre-deciding the result
The article is written from the evidence record rather than from a predetermined conclusion.
Where the evidence supports a statement, we state it.
Where the evidence is weaker, the language should reflect that.
Where the record remains incomplete, the uncertainty remains in the article.
Handling Documents and Data
Chain of custody
Where physical material is relevant and provenance matters, we record what is known about where it came from and how it has been handled.
We do not imply a complete chain of custody where the record does not establish one.
Redactions and personal information
We do not publish private or sensitive information merely because it appears in a source.
Redactions may be retained or applied where disclosure would create unnecessary harm and there is no clear public-interest reason to publish the information.
Images, diagrams and maps
Images should identify their source where reasonably possible.
Illustrations, reconstructions and generated explanatory graphics should not be presented as original documentary evidence.
Captions should make clear what the reader is looking at and whether the image has been cropped, annotated or otherwise altered in a way that matters to interpretation.
Transcription
Where text is transcribed from a scan, photograph or difficult document, the transcription should be checked against the original.
Uncertain or illegible wording should not be silently guessed.
Where a disputed reading matters to the investigation, the underlying passage should be made available where possible so that readers can assess it themselves.
Witness Material
Witness testimony can be valuable, but memory is not a recording.
We consider factors such as:
- how much time passed before the account was recorded;
- what the witness could realistically have seen or known;
- whether later information may have influenced the account;
- whether different versions of the statement exist;
- whether the account agrees with physical or documentary evidence; and
- whether practical factors such as distance, visibility, timing or logistics affect what was possible.
A changed account is not automatically evidence of dishonesty.
Where versions differ, we identify the difference and consider whether the available record explains it.
Interpretation, Hypotheses and Speculation
Documented fact, inference and speculation are not interchangeable.
Where we infer something from the evidence, the wording should make that clear.
A plausible mechanism is not treated as a documented mechanism unless the sources establish it.
A missing record is not treated as proof that the record never existed.
A sequence of events is not treated as proof that one event caused another.
A research hypothesis remains a hypothesis unless the evidence establishes it.
What We Still Do Not Know
Every major Veriarch investigation ends with What we still do not know.
This is not a summary or conclusion.
It records the questions that remain unresolved after the available evidence has been examined.
Some investigations close those gaps later.
Others do not.
Where the record remains open, we leave it open.
Corrections and Updates
If we identify a material factual error, we correct it.
Significant corrections should include a visible note explaining what changed and why.
We do not silently alter a material claim simply to make an earlier version disappear.
Articles may also be updated when:
- new primary material becomes available;
- an archive releases additional records;
- a source previously thought unavailable is located;
- a disputed point is resolved; or
- new evidence materially changes the investigation.
Not every minor wording or formatting change requires a correction notice.
Sources and Citations
Readers should be able to identify the evidence supporting the important factual claims in an investigation.
Where possible, we link directly to:
- original documents;
- archive catalogue entries;
- official reports;
- primary datasets;
- academic papers;
- court or parliamentary records; and
- other underlying sources.
Where a primary source cannot be made publicly available, we identify it as precisely as the circumstances allow.
A secondary source is not treated as independent confirmation when it is merely repeating another source.
AI and Research Tools
Veriarch uses software and research tools to assist with parts of the editorial process.
These tools may help with tasks such as:
- organising research;
- comparing documents;
- checking claim-source relationships;
- identifying inconsistencies;
- reviewing structure;
- transcription support;
- drafting assistance; and
- editorial quality control.
AI-generated output is not treated as evidence.
A claim does not become reliable because a research tool, search engine or language model agrees with it.
The underlying source record remains the basis for factual claims.
Final editorial responsibility remains with Veriarch.
Independence and Funding
Veriarch controls its own editorial decisions.
No donor, advertiser, affiliate partner or outside organisation has the right to determine the outcome of an investigation.
Reader support may help cover costs such as:
- hosting;
- archive fees;
- document access;
- research tools;
- specialist services; and
- maintaining the site.
Where funding is tied to a specific records request or other identifiable research cost, we may explain what the support funded.
Financial support does not buy a finding.
Editorial Identity
Veriarch uses a collective editorial byline.
The wider research team remains private, while the site itself is operated by Rob Hutchison in the United Kingdom.
The collective byline reflects the collaborative nature of the research and editorial process.
Readers should not have to rely on the reputation of an individual author to decide whether an investigation is sound.
The source trail, methodology, corrections and unresolved questions should allow the work to be examined independently.
Your Part In This
Readers are part of the checking process.
If you find:
- a factual error;
- a stronger primary source;
- a missing document;
- a conflicting record;
- a transcription problem;
- a relevant photograph or dataset; or
- specialist information that materially affects an investigation,
please contact us.
If you are submitting sensitive material or your identity requires protection, say so before sending the material itself.
Where a reader identifies a material error, we will correct it.
The Standard
A Veriarch investigation should allow a reader to distinguish between:
- what the record establishes;
- what the record suggests;
- what remains disputed; and
- what is still unknown.
We do not aim to make uncertainty disappear.
We aim to make the evidence easier to inspect.