A reproducible fact-check template

A reproducible fact check lets another careful reader start from the same claim, open the same evidence, understand each transformation and judgment, and see why the verdict follows. Reproducibility does not require that every reader agree; it requires that hidden steps are not doing the work.

The template below is designed for web claims, images, quotations, accounts, records, and timelines. Copy it before research begins so the evidence log is created during the work rather than reconstructed afterward.

1. Claim card

CHECK ID:
WORKING TITLE:
RESEARCHER:
REVIEWER:
OPENED AT (UTC):
STATUS: triage | researching | review | published | corrected

EXACT CLAIM:
CLAIMANT OR CIRCULATING SOURCE:
ORIGINAL URL / POST ID / DOCUMENT ID:
FIRST OBSERVED AT (with timezone):
PUBLIC-INTEREST REASON:
POTENTIAL HARM FROM ERROR:

CLAIM TYPE:
[ ] existence      [ ] identity       [ ] quotation
[ ] image/video    [ ] date/location  [ ] quantity
[ ] legal/policy   [ ] causal         [ ] other

FACT-CHECKABLE COMPONENTS:
C1.
C2.
C3.

OUT OF SCOPE:
DECISION STANDARD:

Quote the claim exactly before paraphrasing it. Preserve material qualifiers such as “all,” “never,” “official,” “caused,” or “on Tuesday.” If a post mixes fact and opinion, identify only the factual components that public evidence can test.

Explain why the check matters. Reach, urgency, safety, and public consequence are legitimate selection factors. Disagreement with the speaker is not.

2. Preservation record

LIVE URL ENTERED:
FINAL URL AFTER REDIRECTS:
RETRIEVED AT (UTC):
PAGE TITLE / ACCOUNT / PUBLISHER:
VISIBLE PUBLICATION OR UPDATE DATE:
CANONICAL URL:
ARCHIVED URL AND CAPTURE TIME:

ORIGINAL FILES:
- evidence ID | filename | format | bytes | SHA-256

WORKING COPIES AND TRANSFORMATIONS:
- source evidence ID | crop/OCR/redaction/conversion | output file | SHA-256

CAPTURE LIMITS:

Keep screenshots in context and retain original downloads separately from crops, annotations, transcripts, or redactions. Verify that archived pages actually contain the relevant material. A hash can help detect later byte changes; it does not prove a file's author, date, or truth.

3. Source plan

Before collecting links, write what direct evidence should exist:

COMPONENT | BEST AVAILABLE PRIMARY SOURCE | SECONDARY CONTEXT NEEDED
C1        |                               |
C2        |                               |
C3        |                               |

SEARCH UNIVERSE:
- jurisdictions:
- date range:
- source languages:
- official systems / archives:
- known coverage gaps:

STOP CONDITIONS:
- sufficient evidence for the decision standard:
- unsafe, intrusive, or prohibited methods:
- deadline and what will remain unresolved:

The best primary source is relative to the component. A recording can establish spoken words; a current regulation can establish operative text; a dataset and methodology can establish reported measurements. None automatically proves a broader interpretation.

4. Evidence log

Use one row for every item that materially changes the analysis:

| Evidence ID | Source, creator, and stable locator | Date and version | Supports | Undermines | Independence | Limits | | --- | --- | --- | --- | --- | --- | --- | | E1 | URL, document ID, page, or timestamp | Original and normalized date | C1, C2, or C3 | C1, C2, or C3 | Direct, copied, shared origin, or independent | Missing context, editable metadata, partial coverage, bias, or uncertainty |

Do not count search-result snippets as evidence when the underlying page is available. Follow citations backward. Two articles using the same press release are one evidence chain, not independent corroboration.

The International Fact-Checking Network's standards require significant sources to be identified so readers can replicate the work, suitable primary sources to be used when available, and relevant evidence on both sides of the claim to be presented. Use that as the minimum discipline even when the result is not a formal newsroom fact check.

5. Source analysis

For each decisive item, answer:

EVIDENCE ID:
WHAT IS THE ITEM?
WHO CREATED OR ISSUED IT?
WHY COULD THEY KNOW?
WHAT WAS ITS PURPOSE AND AUDIENCE?
IS THIS THE RELEVANT VERSION AND STATUS?
HOW DID THIS COPY REACH US?
WHAT TRANSFORMATIONS OCCURRED?
WHAT EXACTLY DOES IT ESTABLISH?
WHAT DOES IT NOT ESTABLISH?
WHAT SOURCE COULD CHALLENGE IT?

Primary sources still require context. The Library of Congress recommends examining a source's creator, purpose, audience, perspective, and agreement or conflict with other sources. The National Archives' analysis sequence—meet the document, observe its parts, make sense of it, then use it as evidence—is a useful review order for more than historical records.

6. Identity, media, and timeline checks

Use only the modules the claim needs:

IDENTITY
- exact names, handles, account IDs, record IDs:
- direct ownership or identity anchors:
- namesake / impersonation candidates kept separate:
- contradictions:

MEDIA
- source copy and derivative relationship:
- reverse-image / keyframe queries:
- creator or publisher chain:
- crop, edit, synthesis, provenance, and context findings:

TIMELINE
- event time:
- capture / upload / publish / update / archive times:
- original timezone and normalized UTC values:
- earliest/latest supported window:
- sequence constraints and impossible overlaps:

LOCATION
- claimed precision:
- independent visual or documentary anchors:
- alternative locations tested:
- sensitive details withheld:

Never merge people on name alone or identify an unknown private person by face. Do not use password recovery, breached credentials, pretexting, private-account probes, or evasion of access controls.

7. Negative-evidence test

EXPECTED EVIDENCE IF CLAIM IS TRUE:
WHO WOULD CREATE IT AND WHY:
COVERAGE OF CHECKED SOURCES:
DETECTABILITY OF OUR SEARCH OR TEST:
RETENTION / SEALING / DELETION / INDEXING LIMITS:
REASONABLE IDENTIFIER VARIANTS CHECKED:

NEGATIVE RESULT MEANS:
[ ] no evidence located
[ ] insufficient evidence
[ ] some evidence of absence within scope
[ ] strong evidence of absence within scope

Do not jump from an empty result to “false.” A negative result has weight only when the expected signal would probably exist and the checked method could probably find it.

8. Analysis by component

Write the reasoning before choosing the headline verdict:

C1 FINDING:
- supporting evidence:
- challenging evidence:
- source-quality judgment:
- remaining alternatives:
- confidence:

C2 FINDING:
- supporting evidence:
- challenging evidence:
- source-quality judgment:
- remaining alternatives:
- confidence:

C3 FINDING:
- supporting evidence:
- challenging evidence:
- source-quality judgment:
- remaining alternatives:
- confidence:

Keep observation and inference separate. “The archived page contains this sentence by 10 June” is an observation. “The author probably added it after criticism” is an inference requiring its own evidence.

Apply the same standard to evidence that favors and disfavors the claim. If a source has a relevant interest or relationship, disclose it; do not discard the evidence solely because the source has a perspective.

9. Verdict and confidence

Choose a label defined before the check:

| Verdict | Use when | | --- | --- | | Supported | The material components are established by appropriate direct and corroborating evidence, with no unresolved conflict that changes the result. | | Partly supported | Some material components hold and others are wrong, overstated, or unresolved. | | Misleading context | The underlying item may be authentic, but framing, omission, date, place, identity, or comparison materially changes its meaning. | | Contradicted | Reliable evidence directly conflicts with the specific material claim. | | Unsupported | Adequate support was not supplied or located, but the available evidence does not establish the opposite. | | Unresolved | Source access, identity, coverage, or conflicting evidence prevents a responsible conclusion. | | Not fact-checkable | The proposition is opinion, prediction, value judgment, or too undefined to test without rewriting it. |

Then publish:

VERDICT:
CONFIDENCE: high | moderate | low
ONE-SENTENCE ANSWER:

WHY:
1.
2.
3.

WHAT WOULD CHANGE THE VERDICT:
LIMITATIONS:
REVIEWER DECISION AND DATE:

Avoid a false precision score. Confidence depends on source directness, independence, coverage, identity continuity, and the size of unresolved alternatives—not the number of links in the bibliography.

10. Publication and correction record

PUBLISHED URL:
PUBLISHED AT (UTC):
SOURCES VISIBLE TO READERS: yes | exceptions explained
METHOD LINKED: yes | no
CONTACT / APPEAL PATH:
NEXT REVIEW TRIGGER:

CORRECTION LOG
- UTC date | requested by | evidence supplied | decision | visible change

Make the sources and method visible enough that readers can reproduce the material steps. Explain any source detail withheld for safety. Invite specific corrections and preserve the earlier conclusion when a material update changes the verdict.

The IFCN calls for a visible, open corrections policy and transparent changes that reach readers of the original. A silent edit repairs text but not the evidence trail.

For capture mechanics, use how to preserve and document a web source. For deciding how to word a blank search result, read what “no evidence” means.