Five investment theses with the kill criteria written down upfront. An AI pipeline reads mechanism news and price action every weekday morning and moves a 0–100 conviction score. Evidence in, conviction out — no vibes.
Every weekday at 07:00 CET a cron pulls prices, scans fresh headlines for each mechanism signal, and makes one LLM call per thesis (Grok) that scores every signal −2…+2 against the thesis mechanics.
Signal scores move the 0–100 conviction. Mechanism/news signals weigh 3× price momentum, and the arithmetic is deterministic pipeline code — the model judges evidence, it never does the math.
Kill and double-down criteria were written before the positions, and are shown verbatim on every thesis page. 5 straight days under 35 → kill review. 5 straight days over 75 → double-down review.
A public experiment in accountable AI research. Nothing here is investment advice or a recommendation to buy or sell anything.