buchiang.com aggregates three orthogonal real-time signals into one terminal:
BTC large-position monitoring (Hyperliquid liquidation map) and a Fear-&-Greed banner sit on top, because macro context changes the read of every signal underneath.
Most prediction-market analysis has the same two failure modes: (1) it's behind a paywall, so you can't audit the track record, and (2) the verdicts are stated only when they're right. buchiang takes the opposite stance: every verdict, every wrong call, every settled outcome stays public, with timestamps. Accuracy is computed live from the same data the page renders. There is no "select cases" presentation.
If the 30-day accuracy drops, you see it on the dashboard before anyone has time to spin it.
AI Jury inference happens on dedicated hardware separate from the public web tier — the inference layer pushes verdicts via signed HMAC contract; the public tier never has access to the models or thresholds. CEX scanning is a separate cron job on its own machine. Three isolated stacks, three separate concerns, one published surface.
This separation is why the system can stay running while individual components are rebuilt.
Found a bug, have an improvement idea, think a winrate looks wrong, want a feature — all welcome:
info@buchiang.com
No support SLA, but every email gets a real reply. I'd rather spend time answering mail and shipping improvements than running a Telegram group or Discord.
Don't want to come back and refresh every day:
/rss/verdicts.xml — every AI Jury verdict pushed in real time/rss/whales.xml — every Polymarket whale signal pushed in real timeSubscribe with any RSS reader (Feedly / NetNewsWire / Reeder / Inoreader). Zero tracking, zero account, zero client-side overhead.
[ buchiang@v1 // online since 2026-Q1 ]