Methodology
Our investigation approach. Every verdict shows its evidence chain.
Layer 1: Existing Fact-Check Lookup
We check the Google Fact Check Tools database — a global index of IFCN-certified fact-checkers. If AFP Fact Check, Full Fact, PolitiFact, Snopes, or BOOM have already reviewed this claim, we surface their findings immediately.
Layer 2: Bilingual Source Search
We search in both Bengali and English across trusted sources — international wire services (Reuters, AP, AFP), Bangladesh national outlets, government sources, and fact-checking organisations.
Layer 3: Absence Analysis
When little evidence is found, we analyse why. Did the event never happen? Is it too recent for coverage? Is the claim deliberately vague? Understanding the absence of evidence is as important as the evidence itself.
Layer 4: Image Forensics
When images are submitted, we analyse them for signs of manipulation — editing artifacts, inconsistent lighting, text overlays, social media formatting inconsistencies, and AI-generated image characteristics.
Layer 5: Pattern Detection
We check for common misinformation patterns: urgency language, vague attribution, recycled content from previous events, emotional manipulation, and tribal framing.
Human Review
AI investigates. Humans decide. Every verdict requires admin review before publication. No fact-check is auto-published without human confirmation.
Verdict Categories
We use 14 verdict categories aligned with international fact-checking standards (IFCN, ClaimReview, PolitiFact, Snopes, AFP). Each verdict is assigned based on the weight of evidence and confirmed by a human reviewer before publication.
Verified (সত্য)
The claim is accurate. Multiple independent, credible sources corroborate the core assertion. No material omissions or distortions were found. This is the highest confidence rating — reserved for claims where the evidence is clear and unambiguous.
Mostly True (আংশিক সত্য)
The claim is largely accurate but needs clarification, additional context, or contains minor inaccuracies that do not change the overall conclusion. The core assertion holds up under scrutiny, but the way it is presented could give a slightly incomplete picture. Used by PolitiFact and Snopes for claims that are substantively correct with small caveats.
Half True (আধা সত্য)
The claim contains elements of both truth and falsehood. Important context is missing, or the claim cherry-picks data to support a conclusion that does not tell the full story. The statement is not outright wrong, but it paints a materially incomplete or skewed picture. This is one of the most commonly applied ratings across global fact-checkers.
Misleading (বিভ্রান্তিকর)
The claim contains a kernel of truth but presents it in a way that distorts the reality. This could involve selective quoting, misleading statistics, false equivalences, or framing that leads the audience to an incorrect conclusion. The distinction from Half True is that the distortion appears deliberate or the framing fundamentally changes the meaning.
Mostly False (অধিকাংশ মিথ্যা)
The claim contains a small element of truth but ignores critical facts that would give a completely different impression. The overall conclusion is wrong. The grain of truth may be a real statistic taken wildly out of context, a real event twisted beyond recognition, or a factual premise used to support a baseless conclusion.
False (মিথ্যা)
The claim is not accurate. Credible sources directly contradict the assertion with evidence. There is no significant element of truth. This rating is applied when the evidence clearly and unambiguously shows the claim to be wrong.
Fabricated (বানোয়াট)
There is no evidence that the claimed event, statement, or scenario ever occurred. The content shows signs of being entirely invented — fake screenshots, non-existent sources, invented quotes, or AI-generated content presented as real. This is distinct from False (which debunks a real claim) and Altered (which modifies real content).
Altered (পরিবর্তিত)
Real photo, video, or audio that has been significantly doctored, edited, spliced, or digitally manipulated to change its meaning. The source material is genuine but has been tampered with — cropped to remove context, edited to change text, spliced to create fake sequences, or processed through AI tools to alter faces or voices. Distinct from Fabricated (entirely invented) and Miscaptioned (real media, wrong description).
Miscaptioned (ভুল ক্যাপশন)
The photo, video, or media is real and unaltered, but the caption, date, location, or attribution is wrong. The content is being shared with a false description — a real photo from one country labelled as another, a years-old image presented as current, or a real quote attributed to the wrong person. The media itself is authentic; only the framing text is false.
Out of Context (প্রসঙ্গবিহীন)
The content is real and accurately reproduced, but it has been removed from its original context in a way that changes its meaning. This includes clipping a longer statement to reverse its intent, sharing a genuine document without the surrounding discussion, or presenting a conditional statement as absolute. Distinct from Miscaptioned (wrong label on real media) in that the framing distortion is structural, not just descriptive.
No Evidence (প্রমাণ নেই)
After extensive searching across multiple sources in both Bengali and English, we found no credible evidence to either confirm or deny the claim. This is not the same as False — it means the claim is currently unverifiable. The claim may be too recent, too obscure, or deliberately constructed to be unfalsifiable.
Satire (ব্যঙ্গ)
The content originates from a satire publication, parody account, or comedic source that was not intended to be taken as factual reporting. The original creator intended it as humour or social commentary, but it has been shared without that context, leading people to believe it is real news.
Scam (প্রতারণা)
The content is part of a phishing scheme, financial fraud, fake giveaway, impersonation scam, or other deceptive scheme designed to steal money, personal information, or account access. Common patterns include fake government payment portals, celebrity endorsement scams, lottery or prize notifications, and job offer fraud. Particularly prevalent on Facebook and WhatsApp in Bangladesh.
Outdated (মেয়াদোত্তীর্ণ)
The claim was accurate at one point but is no longer true due to changed circumstances — updated laws, revised statistics, policy reversals, or new scientific evidence. The original information was correct when first reported but is now being recirculated without acknowledging that the situation has changed.
Historical Distortion (ইতিহাস বিকৃতি)
Documented historical events are deliberately misrepresented by inversion, attribution-swap, chronological reordering, or erasure, where the authoritative historical record is definitively clear and the distortion serves a political or ideological narrative. Distinct from False (which debunks a specific factual claim) in that the distortion targets the shared historical record itself. Distinct from Fabricated (which invents events) in that the events are real but their meaning, cause, or agent has been reassigned. Common in political discourse where historical memory is contested.
Corrections Policy
We take accuracy seriously. If you find an error, see our corrections policy.