1. Normalise the claim
Evidrai separates the checkable factual claim from rhetoric, opinion, framing, and repetition so the assessment is anchored on something specific.
About Evidrai
Built to make evidence easier to inspect.
Evidrai is an early-access verification platform for people who need to understand whether a claim is supported, contradicted, missing context, or still unproven.
What Evidrai is
Evidence assessment, not fact-checking theatre.
The aim is not to produce a magic truth label. The aim is to expose the evidence trail clearly enough that a human can decide what to trust, what to question, and what needs more work.
- Evidrai is decision support, not an oracle.
- YouTube URL extraction is best-effort; pasted transcripts are the reliable fallback.
- Evidence can be incomplete when sources are unavailable, blocked, paywalled, or newly developing.
- Confidence reflects the reviewed evidence, not absolute certainty.
Evidence methodology
Trust is earned by showing the work.
Evidrai produces transparent, evidence-based assessments that explain what is supported, what is contested, and where confidence is limited.
2. Gather relevant evidence
The system looks for sources that can corroborate, contradict, contextualise, or weaken the claim. More sources do not automatically mean stronger evidence.
3. Classify source role and quality
Sources are grouped by how they relate to the claim, with attention to credibility, proximity to the evidence, transparency, and whether the source is primary, expert, institutional, or secondary.
4. Compare evidence, not volume
Evidrai weighs corroboration, contradictions, missing context, and caveats. Repetition across low-quality sources is treated differently from independent supporting evidence.
5. Produce an inspectable assessment
The verdict, confidence, caveats, reasoning, and source trail are shown together so users can see why the system reached its assessment and where uncertainty remains.
How Evidrai scoring worksSource score, evidence strength, confidence
Source score0.0-5.0
How strong an individual source is for the specific claim.
Evidence strength0-10
How strong the reviewed evidence set is after support, contradiction, and repetition are considered. For false or contradicted claims, this is shown as contradiction strength.
Confidence0-100 signal
How confident the system should be in the verdict, given source quality and uncertainty.
Authority30%
Is this source authoritative for this claim type? Primary records, official datasets, filings, transcripts, and direct evidence carry more weight than commentary.
Relevance25%
Does the source directly address the exact claim, or is it only loosely related background?
Directness20%
Is the source close to the underlying evidence, or is it repeating what another source claimed?
Recency10%
Is the source temporally appropriate? Current claims need current evidence; historical claims may need contemporaneous records.
Independence10%
Does this source add an independent evidence chain, or is it amplifying the same report, briefing, post, or wire story?
Bias risk5%
Does the source have a direct incentive to frame the claim selectively? Bias risk is treated as a modifier, not a veto.
Source score bands
4.5-5.0Very strong source3.75-4.49Strong source2.75-3.74Useful or mixed source1.75-2.74Weak source0-1.74Poor, indirect, or irrelevant source
Guard rails
- Primary evidence carries more weight than repetition.
- Five articles based on the same briefing may count as one evidence chain.
- Context helps explanation, but should not inflate confidence.
- Strong contradictions reduce confidence in the claim, even when it is widely repeated.
- When a claim is rejected, Evidrai should say “claim unsupported; credible contradiction found” rather than implying the disproof itself is weak.
- No score is shown without an explanation path.
Evidence over repetitionConfidence is not certaintyReasoning remains inspectableCaveats are surfaced