AI Forecasting Journal

You'll build a personal forecasting journal that runs locally on your machine. Log what you think will happen, lock in a probability, resolve it when the answer comes in. A calibration plot shows your predicted probabilities against actual outcomes so you can see where you're overconfident or underconfident.

One-click PNG export gives you a 1200×628 share card ready for LinkedIn or Twitter, with your Brier score and calibration shape but never your individual predictions. That's the viral loop, and it's the reason the spec pays for itself if you're building a reputation as a calibrated thinker.

Ships with 15 curated AI-governance starter questions (compute thresholds, EU AI Act, RSP triggers, AGI claims, incident severity) so you're not staring at an empty page on day one.

Everything lives in your browser's localStorage. No account, no server, no analytics. Your predictions about specific labs, specific policies, specific capability thresholds stay where they belong, which is on your laptop.

The whole thing builds in about two hours on Claude Code free tier.

Forecasters, AI policy researchers, think tank analysts, and anyone who wants to log sensitive predictions somewhere that isn't a server.