I built Acuity because I was tired of fact checkers that only focus on true/false data points while ignoring the manipulation embedded in the structure of the text.
We know that a story can be factually accurate but structurally dishonest (like using zombie facts from 2022 to imply a crisis in 2025, or using higharousal emotional language to force a behavioral response).
Acuity is a forensic analysis engine that scores content (0-100) based on three specific vectors: 1. Reality Anchoring: Does it cite existent sources? (We use a "Freshness Protocol" to handle breaking news latency). 2. Tribal Engineering: Does the text use In Group/Out Group framing to bypass critical thinking? 3. Intent Analysis:Is the language Descriptive (Journalism) or Prescriptive (Commanding/Coercive)?
The tech stack: - Core: Python (FastAPI) on Render. - Intelligence: A hybrid pipeline using Grok (for unmoderated structural analysis) and Tavily (for real-time swarm verification). - Scraping: We implemented a pincer movement for ingestion: - Desktop: A Chrome Extension (Manifest V3) using activeTab to read DOM text directly (bypassing blocking). - Mobile: A React Native (expo) app that integrates into the native iOS/Android Share Sheet. - Hard Targets: We use Firecrawl to handle sophisticated anti-bot countermeasures when serverside scraping is required.
The hardest problem: Mobile distribution was a nightmare. We realized users wouldn't copypaste URLs. We ended up building a native Share Extension that allows you to "Share" a paywalled article from Safari/WSJ directly to Acuity. On iOS, we use the NSExtensionJavaScriptPreprocessingFile to extract the text from the active webview, allowing us to analyze paywalled content without breaking encryption or logging ineffectively giving the user "xray vision" for their own screen.
It's currently in Alpha. I’m not selling user data (the business model is B2B data licensing for AdTech later, not consumer surveillance).
I’d love feedback on the scoring logic specifically if you find false positives where it flags legitimate opinion pieces as manipulation.
Thanks!