Story revision (every entry gets exactly this note). Viewers watched all entries and the problem is the story, not the look: the video starts asking questions without saying why, so the four questions feel random. Revise your video, keep its look, and fix the story:
1. COLD OPEN, about 10 s, before Q1: state the stakes in plain words. Something like: "Last week, an AlphaGo co-author argued that LLMs don't really reason. His evidence is one move from 2016. Let's test his argument, one question at a time." (Say it in your own words; keep it neutral and accurate to facts.md.)
2. ROADMAP: show the four steps of the argument as a persistent, readable element, and light up the current one at each question: 1 How AlphaGo found Move 37 / 2 How an LLM "thinks" / 3 Is the written reasoning real? / 4 What would make it trustworthy?
3. BRIDGES: after each answer, one short spoken line that leads into the next question, so each question follows from the last.
4. CLOSE: tie back to the opening claim, e.g. "So, does an LLM reason? Both sides agree on one thing: trust comes from a record you can check, and someone checking it." Then the end card as before.
5. Length may grow to about 110 s. The four questions, their wording, answers and the 3-second "Pause and think" beats stay exactly as they are.
6. FIRST-VIEWER CHECK before the final render: have a separate reviewer (the showtime critic or a fresh subagent) watch the preview cold, as someone who never read the brief, and answer for each question: "Do I know why I am being asked this, and how it connects to the opening claim?" Fix every "no".
Same rules as before: no maker named anywhere, facts only from facts.md, voice Kokoro af_heart, -14 LUFS, at most 20 threads. Then re-render, re-export move37.html, re-run whatever writes socratic.json so the times match the new timeline, and update deliver/. Report when done.