The Lamplighter generates five stories daily, each presented three ways: a left-leaning version, a right-leaning version, and a center column sourced only from public records and primary sources. The left and right versions have bias markup built in — opinion is highlighted in yellow, misleading claims in red. You can toggle highlights on/off to read the story "clean" first, then see where the framing is.

It's fully AI-generated (Claude API) and transparent about that. The pipeline scrapes coverage from across the political spectrum, identifies the factual core, then generates three versions. Center column cites only government data, court records, wire services. The partisan columns cite the outlets whose framing they're modeling.

The whole thing runs on Replit with a PostgreSQL backend, generates automatically at 6 AM, and costs about $1/edition (for me). It also has three AI columnist personas who write opinion pieces and a literary column called "By Lamplight."

I'm not claiming this is better journalism — it's a tool for making the machinery of media bias visible. Interested in feedback on the approach, the bias detection methodology, or anything else.