Two Lenses

Two Lenses: one story, every lens. Live and running at https://two-lenses.wajohack.workers.dev/launch/ Most people see one outlet’s version of a story. On October 24, 2026, police counted 400,000 people on the streets of France. Fox News called them “rioters”; NBC said the protests were “for the most part peaceful and large.” You only notice choices like these when you see both reports. Two Lenses takes one pasted news video and finds other outlets’ coverage of the same event. Gemini 4 Argon watches every report in full and lines up the moments that answer the same question: the first words, the stated cause, who gets to explain it, how bad it was, and why viewers should care. Each moment shows both clips side by side with a named framing technique, a verbatim quote and its timestamp. Then the loop takes over. The explainer is rendered and recorded in headless Chrome, and Argon watches its own recording alongside the sources. It critiques the result for grounding, fairness and readability, together with deterministic checks against the captions, then fixes the explainer and re-renders until it passes. Fairness is a first-class criterion: if one outlet gets scrutiny and the other doesn’t, that’s an error. In our first run, Argon’s critic caught that its own analysis had written 1:05 as “105 seconds”. It explained the mistake and moved every affected clip back to where the words are actually said. Grounding went from 0.55 to 0.90. It’s built for anyone stepping outside their media bubble, and for teachers and librarians running media-literacy classes who need real, current, timestamped examples. It’s open source (MIT).

Team