Guardian AI Scorecard v1.0

Grok has native, real-time access to the full X firehose — not a sample, not a delayed feed. Guardian sees every trending topic, every engagement spike, and every viral post as it happens. This is the same live intelligence layer that powers WeBetAI's sports picks pipeline, repurposed for narrative analysis. No other consumer-accessible AI has this depth of real-time social data.

Multi-signal authenticity scoring: engagement velocity (organic growth vs. sudden spike), account diversity (unique voices vs. coordinated repetition), source origin tracing (primary source vs. circular citation), and sentiment distribution (natural range vs. manufactured consensus). Each signal is scored independently before the composite authenticity rating is computed. The system doesn't decide what's true — it measures how authentically a narrative is spreading.

Guardian doesn't just flag individual posts — it maps how narratives form, propagate, and evolve across the platform. Grok's native understanding of X's conversation graph means Guardian can trace a story from its origin point through amplification waves to mainstream adoption. This structural analysis distinguishes organic cultural moments from manufactured campaigns at the narrative level, not the post level.

Every trending topic is cross-referenced against primary sources: official data releases (BLS, Fed, SEC filings), verified news wire APIs, and original source documents. When a claim goes viral, Guardian checks whether the underlying source actually says what people claim it says. This catches the most common form of inauthentic content — real sources misrepresented, statistics stripped of context, or headlines that contradict their own articles.

Interest profiling based on engagement history: which topics you interact with, which challenges you accept, which digest items you click through. Over time, your daily digest reflects what makes you interesting — not what makes you angry. The anti-algorithm: optimized for relevance to your actual life, not for time-on-screen. Sports, tech, markets, culture, science — weighted to you.

Authentic stories with clear, verifiable outcomes are automatically converted into friendly P2P challenges. The system identifies the "bettable moment" in a narrative — the specific claim that can be resolved with time. This is the conversation engine: a $5 challenge between friends on something real is the most natural on-ramp to an in-person conversation that exists.

SMS-first delivery via Twilio. No app download, no account creation, no feed to scroll. Phone number is the identity. The daily digest arrives like a text from a friend — 5 authentic stories, scored and summarized, with a challenge you can forward in one tap. The medium is the message: if it arrives by text, you engage by text. If it arrives in a feed, you scroll. WeBetAI chooses text.

Every design decision is evaluated against a single question: does this reduce screen time or increase it? The digest is intentionally short (5 items). There is no infinite scroll. There are no push notifications beyond the morning digest. The success metric is time-to-put-phone-down, not time-on-platform. Guardian succeeds when you're not using it.

Content is analyzed for emotional engineering patterns: fear triggers, urgency fabrication, outrage framing, and false binary construction. Grok's language model identifies when phrasing is designed to bypass rational evaluation and provoke an emotional response. This scoring layer surfaces content that is factually accurate but emotionally manipulative — the hardest category to detect and the most damaging to mental health.

Tracking infrastructure is built: which digest items users engage with, which challenges are accepted vs. ignored, which topics drive real conversations. However, this data does not yet feed back into the authenticity model or personalization weights. The architecture is ready but requires ~30 days of user interaction data before the feedback loop can be activated. This is the bottleneck to reaching 95th composite.