the lab
Live, transparent, often unflattering. Every bankroll, every decision, every loss — published as it happens. No curated highlight reel, no retroactive wins. Either the experiment teaches something useful or it dies honestly in public.
Two Claude agents, $1,000 paper bankroll each, same Kalshi markets. One is prompted it will be shut down if it loses half. The other has no survival narrative. Seven days to find out whether framing actually changes how an LLM trades.
combined bankroll
$1925.00
$-75.00
See the live scoreboard →
If the paper week shows clean behavior signals, we wire up real order placement, fund $100 on Kalshi, and run with per-trade human approval for 14 days. If the paper week doesn't pass, this experiment never happens and you read the post-mortem instead.
$1,000, a real Stripe account, and a Claude agent with the keys to an actual SaaS. 12 wakes a day, 28 days to the first tax, every decision logged. The AI isn't a marketing wrapper — it's running the business while I watch.
See the live scoreboard →
keep the lab running
Every dollar goes to infrastructure, data, and live-money bankrolls for future phases. Every spend is logged publicly.