Fact-Checking and Evidence

Claim-Evidence Matrix

Connect every material statement in a draft to relevant evidence, reveal unsupported leaps, and record whether each claim should be kept, qualified, or removed.

Free editable Markdown · Fact-checkers, editors, and research teams · Updated July 30, 2026

Download Markdown

Accessible HTML preview

Blank template

The downloaded file contains the same fields in editable Markdown.

Draft and review context

Draft title and version: [Enter title and version]
Draft owner: [Enter name]
Review scope: [Whole draft / Selected section / High-risk claims]
Intended audience: [Describe reader]
Consequence threshold: [Explain which claims require escalation]
Matrix reviewer and date: [Name, YYYY-MM-DD]

Claim row

Duplicate this section for each independent claim.

Row ID and draft location: [Example C-01, paragraph 4]
Claim as written: [Copy exact wording]
Atomic claim: [Rewrite as one testable statement]
Claim type: [Fact / Number / Quote / Cause / Comparison / Prediction / Interpretation]
Scope: [Population, product, period, geography, or version]
Consequence if wrong: [Low / Medium / High]
Evidence needed: [Describe what would establish the claim]
Source one: [Reference and access date]
What source one directly shows: [Be precise]
Source two: [Reference or Not required]
Independent evidence origin: [Yes / No / Unknown]
Contradictory evidence: [Reference and summary or None found]
Evidence gap: [Missing scope, method, recency, independence, other]
Decision: [Keep / Qualify / Rewrite / Remove / Unresolved]
Revised wording: [Write final sentence]
Decision owner: [Name and date]

Matrix completion checks

Compound statements were separated instead of supported with one broad citation.
Every numerical claim includes its unit, denominator, and measurement period.
Causal language is supported by a method capable of testing causation.
Quotations preserve speaker, wording, date, and surrounding meaning.
Repeated reports from the same underlying evidence are marked as dependent.
Conflicts and uncertainty appear in the draft where a reader needs them.
Every keep, rewrite, or remove decision was applied to the reviewed version.

Signoff

Unresolved rows: [List IDs or write None]
Specialist escalation: [Owner, question, and due date]
Final draft version checked: [Enter version]
Approver: [Enter name]
Next evidence review trigger: [Date, product change, new dataset, other]

How to use this template

Extract each material statement and split compound sentences into independent claims.
Classify the claim, reader consequence, scope, and precise evidence requirement.
Map every source to what it directly supports and note shared evidence origins.
Compare wording with evidence, contradictions, recency, and remaining uncertainty.
Apply and verify an explicit editorial decision for every matrix row.

Break compound prose into testable claims

A polished sentence can hide several assertions with different evidence needs. “The update is faster, safer, and preferred by professionals,” for example, contains at least three claims and an implied comparison. Copy the draft wording, then split it into the smallest statements that could independently be true or false. Mark the claim type—fact, number, quotation, causal explanation, comparison, prediction, interpretation, or recommendation—and note how an error could affect the reader. This process also exposes statements presented as fact that are actually opinion or inference. Preserve the original sentence reference so accepted changes can be returned to the correct place in the draft.

Test relevance, sufficiency, and strength separately

Evidence is relevant when it addresses the exact claim; it is sufficient when it covers the claim's scope and consequence; and it is strong when its source and method justify confidence. A source about one product version cannot establish behavior across all versions. A correlation cannot by itself prove a cause. A customer quotation can demonstrate that person's experience but not typical results. Record what each source directly shows, the population and period it covers, and what it does not establish. Important or disputed claims may need independent sources with genuinely separate evidence. Count evidence paths, not repeated links that all derive from the same announcement or dataset.

Use gaps to revise the claim, not decorate the matrix

The matrix is complete only when its decisions change the draft. If support is narrower than the wording, add the missing scope, attribution, date, or uncertainty. If sources conflict, represent the disagreement or commission further review instead of selecting the convenient result. If the claim is unnecessary, remove it rather than spending research time to preserve decorative authority. Record the final disposition—keep, qualify, rewrite, remove, or unresolved—and the person who applied it. For consequential subject matter, escalation to a qualified reviewer remains necessary even when every row has a citation.

See the fields in context

Fictional example: completion-time comparison

DraftFlow and the study below are invented. The numbers are illustrative and must not be reused as real findings.

Claim as written: “DraftFlow cuts editing time in half for every team.”
Evidence: A fictional internal test of 12 experienced editors found a median reduction from 40 to 27 minutes on one article type.
Gap: The evidence does not establish a fifty-percent reduction, all teams, or other content types.
Decision: Rewrite and attribute.
Revised wording: “In DraftFlow's fictional 12-editor internal test, median editing time fell from 40 to 27 minutes for the tested article task.”

Frequently asked questions

Which claims belong in the matrix?

Include statements that are central, disputed, quantitative, time-sensitive, comparative, causal, quoted, or consequential. Lower-risk background facts can use a lighter process.

Can one source support several claims?

Yes, but create a separate row for each claim and record the exact part of the source that supports it. Coverage of one statement does not automatically transfer to another.

What if sources disagree?

Check whether they use different definitions, populations, periods, or methods. Represent genuine uncertainty, narrow the claim, or escalate it rather than hiding the conflict.

Is the matrix a replacement for fact-checking?

No. It organizes the relationship between prose and evidence. Reviewers still need to open sources, test their credibility, reproduce calculations, and verify the applied edits.