How It Works
Why Bullshit Goes Undetected
Every day, the internet produces roughly blog posts, tweets, and AI-generated texts.
How much of it actually contains new information?
Princeton professor Harry Frankfurt defined it this way:
“Bullshit is more dangerous than lies. You can refute a lie, but bullshit gives you nothing to refute.”
— Harry Frankfurt, "On Bullshit" (2005)
Roast Engine quantifies this problem.
The 5-Dimension Framework
Every piece of communication is precisely measured across 5 independent dimensions. Each scored 0–100.
Cliché
clicheIs it just repackaging the same old ideas? Checked against 12,000+ patterns in our cliché database.
Based on Shannon's information entropy — the more predictable a word sequence, the less information it carries.
Pretentiousness
pretentiousAll sizzle, no steak? Measures buzzword density, modifier-to-fact ratio, and unverifiable claims.
Grice's conversational maxims — detecting violations of the 'quantity' maxim. Excess modifiers are decoration, not information.
AI Slop
ai_slopDoes it sound like bland AI output? Analyzes 10 AI-generation markers.
Perplexity + Burstiness analysis — human text is irregular, AI text is uniform.
Emptiness
emptinessHow low is the information density? Measures verifiable facts per word count.
Information density = verifiable facts / total words. Lower means emptier.
Clickbait
clickbaitDetects curiosity gap manipulation, emotional triggers, and overpromising.
Loewenstein's information gap theory — detecting patterns that deliberately withhold information to trigger curiosity.
The weighted average of all 5 dimensions becomes the Jolong Score.
The Jolong Score
Higher means more BS. Closer to 0 means it's genuine.
Fresh Gourd
0 – 25
Specific facts, verifiable data, and genuine personal experience.
Stale Gourd
26 – 75
Some BS detected but still salvageable.
Rotten Gourd
76 – 100
Unverifiable claims, buzzword soup, information density approaching zero.
How the Analysis Works
From the moment text is entered to the final diagnosis.
Input
Enter text or a URL. For URLs, the content is automatically extracted.
Supports: Korean, English / LinkedIn, blogs, news, job postings, social media, etc.
Preprocessing
Normalizes text, detects language, and splits into sentences.
Metadata extraction: word count, sentence count, modifier ratio, number ratio
5-Dimension Analysis
Specialized analyzers independently score each dimension.
Cliché, Pretense, AI Slop, Emptiness, Clickbait — each 0-100, with evidence
Context Interpretation
Analyzes the context in which the text was written.
Platform (LinkedIn vs news vs social), genre (bio vs report vs marketing), tone (formal vs casual)
Reaction Prediction
Simulates how the world would react to this content.
By generation (Gen Z, millennials), platform (LinkedIn, X, Instagram), culture
Comprehensive Diagnosis
Dr. Park synthesizes all data into a witty, sharp diagnosis.
Overall score + 5-dimension radar chart + highlights + improvement points
We Analyze Society, Not Just Text
74 roast generators out there just analyze text.
Roast Engine interprets human communication through an anthropological lens.
The same words land completely differently depending on who says them, where, and to whom. Roast Engine goes beyond detecting BS — it predicts how the world would react.
4 Layers of Analysis
Which words and structures are cliché?
Cliché DB, buzzword frequency, modifier/fact ratio, sentence structure repetition
What context did this come from?
Platform conventions, genre expectations, social positioning analysis
What's the gap between intent and perception?
Humble-brag detection, emotional manipulation patterns, cognitive dissonance measurement
How is this communication received across cultures?
Korean/American/global context differences, generational sensitivity, humor reception
Reaction Prediction Simulation
The same text gets completely different reactions depending on the audience.
“Satire is a sort of glass, wherein beholders do generally discover everybody's face but their own.”
— Jonathan Swift
Jonathan Swift
1667–1745
아일랜드 풍자 작가. '걸리버 여행기'의 저자. 영문학 역사상 가장 날카로운 풍자가.
The Battle of the Books, 1704
Real Analysis Results
From the most clichéd content to the most genuine. See how the same engine diagnoses them differently.
Dr. Park Jo-Long's ReportSymptoms — 자기소개 예시
저는 혁신적인 리더십으로 조직의 시너지를 극대화하고, 차별화된 고객 경험을 제공하며, 끊임없는 도전과 성장을 통해 세상을 변화시키는 데 기여하고 있습니다.
95
Dr. Park's Comprehensive Diagnosis
한 문장에 경영 컨설팅 회사 브로셔의 모든 키워드를 우겨넣으신 수고에 경의를 표합니다. '혁신적 리더십'부터 '세상 변화'까지, 구체성이라곤 찾아볼 수 없는 완벽한 허공 타격이시네요. 이 정도면 조롱학계의 교과서 사례로 활용하겠습니다.
Dr. Park Jo-Long's ReportSymptoms — 채용공고 예시
자율적이고 수평적인 문화 속에서 함께 성장할 열정 있는 인재를 찾습니다. 우리는 워라밸을 중시하며, 자유로운 근무 환경을 제공합니다.
85
Dr. Park's Comprehensive Diagnosis
전형적인 '좋은 말만 모아놓기' 채용공고입니다. '자율적이고 수평적'이라는 수식어 뒤에 숨은 실제 근무환경, 연봉, 업무내용은 하나도 알 수 없습니다. 이런 공고를 보면 지원자들이 '아, 여기는 아직도 이런 걸로 사람을 구하는구나' 싶겠어요.
Technology Stack
The core technology behind Roast Engine.
Analysis Engine
5-dimension analysis pipeline built on large language models. Context interpretation to diagnosis in one flow.
Cliché DB
210 patterns, 11 categories
Reaction Prediction
3 generations x 3 platforms x 3 cultures. Calculates virality potential and controversy risk. (Experimental)
Learning System
31 lessons, 6 quiz types (concept/T-F/multiple choice/identify/compare/predict) in an interactive curriculum.
Certification
HMAC-SHA256 verifiable certificates. Unique serial numbers with public verification API.
Public API
REST API (v1) — cliché scan + LLM analysis + reaction prediction unified endpoint. (Closed beta)
POWERED BY
Behind every diagnosis, training, and prediction at Roast Lab is Roast Engine. An anthropological AI engine that reads society, not just text.
Try It Yourself