How Nokka turns "is this real?" into receipts.
A look under the hood — the pipeline that takes a suspicious post and produces an evidence-backed, shareable truth card.
The check pipeline
Submit
IngestUser drops a link, screenshot, or claim. We extract the core claim and detect the content type.
Retrieve
EvidenceWe pull candidate evidence: official sources, news, archives, reverse-image and known-scam databases.
Reason
AI verdictAn LLM weighs evidence into a verdict + confidence, and writes the one-line, person-fair explanation.
Review
Human-in-loopLow-confidence or flagged claims route to the human review desk before they go public.
Card it
DistributionThe result becomes a shareable truth card with a permanent public check link.
The responsibility rails
Claims, not people
We rate the claim and explain the receipts. We never identify private individuals or invite pile-ons.
Show the work
Every card carries confidence, source count, and an honest 'what we couldn't confirm'.
Funny, then fair
Comic templates are for low-risk viral nonsense. Sensitive topics default to the serious, evidence-grade template.
System at a glance
Next.js App Router
Routes + RSC
Evidence store
Sources & cards
AI verdict engine
Claim reasoning
Review desk
Human gate
This is a frontend prototype. The pipeline above is mocked with sample data to show the full product experience.