About Rumour Shield
Evidence-centered claim checking for a multilingual, messaging-first society. Understand the claim, examine its sources, and make uncertainty visible.
Our purpose
No traceable evidence, no confident verdict.
Rumour Shield helps people examine public claims before sharing them, especially when misinformation affects civic life. The aim is an understandable assessment backed by evidence, not an answer based on popularity, repetition or an AI model's memory alone.
It supports claims in English, Hausa, Yoruba, Igbo, Arabic and Nigerian Pidgin. A source does not have to use the same language as the person asking: relevant evidence can be assessed in one language and explained in another.
Three principles behind the method
Inspired by Matn analysis, Isnad and narrator evaluation, adapted to modern information verification.
- Matn
Understand the claim
Examine what was actually said. Preserve names, dates, quantities, negation and context, while distinguishing checkable claims from opinions and predictions.
- Isnad
Trace the evidence
Look for the origin of a report and the sources it relies on. Repeated coverage from one origin is not independent confirmation; missing links remain limitations.
- Rijal
Assess the source
Examine reliability, proximity to the event and independence. A familiar publisher or institution does not make every statement it reports correct.
From question to assessment
Meaning is checked before evidence search. If the question cannot be interpreted reliably, the system requests clarification or a retry instead of issuing a verdict.
Receive and prepare
A question or claim arrives as text, a link or supported media. Recognisable private details in the claim text are redacted before verification.
Check the meaning
The engine identifies the language, separates checkable details and reviews its interpretation against the original. It attempts to repair misreadings or asks for clarification when needed.
Retrieve source passages
Search seeks relevant evidence in available languages, including primary records. Retrieved passages and citation links form the basis for checking each part of the claim.
Compare the evidence
Sources are grouped by shared origins and assessed for reliability, independence and timing. Supporting evidence, contradictions and gaps are considered together.
Explain the assessment
Findings are checked against the retrieved evidence. A report explains what is supported and what remains uncertain, with translation into the claim's supported language when available.
Preserve the record
Completed reports include source links and an integrity receipt. Cases flagged for review enter a review queue; that flag does not mean a person has already checked the result.
Safeguards and limits
A useful result should make both its evidence and its boundaries clear.
- Evidence, not certainty
- Results use Confirmed, Rejected or Indeterminate, with findings for individual parts where available. An evidence score is not a guarantee of truth. Missing passages, conflicting accounts and outdated information can limit a conclusion.
- Privacy with limits
- Redaction targets email addresses, recognisable Nigerian phone numbers and certain long numbers in claim text. It cannot catch every private detail. Avoid submitting confidential information: saved reports may be publicly accessible.
- Translation with traceability
- Reports and excerpts can be translated into the claim's supported language, while original source text and links remain available. Translation can fail or alter meaning; a translated passage is not a new, independent source.
- Receipts protect record integrity
- Digital fingerprints identify the saved claim, evidence and report. They help check whether that record has changed, not whether a claim is true. Receipts currently use a local ledger, not public blockchain notarization.
Access and availability
The web app is the main entry point. WhatsApp and Telegram integrations use the same verification pipeline when configured. Live search, media processing and translation depend on the services enabled for a deployment.
Use with judgement
Rumour Shield is under active development. Automated analysis and human review can both make mistakes. Inspect the evidence behind important decisions; this tool is not an emergency service or a final authority.