We build an agentic system that reads classic novels into a library of story parts and that plans and writes long audio serials. We publish what we learn as short notes, one for each finding. A failure gets a note too. A larger finding becomes a technical report. Each report asks one question. It gives its method at a fixed commit, its results with their scope and the results that failed. A fresh copy of the code at that commit must give the same numbers before we publish a report. Two referees read it first.
Our first report is in preparation for November 2026.
Each report uses only novels from before 1931 and our own pages. A correction is a new version. We keep each version at its own address.