Strategy lab

Every idea is backtested on real prices (closing odds, Kalshi history or our own Kalshi logs) with costs and fees included. Filters are chosen on TRAIN data and judged only on later, unseen TEST data. Promising = profitable in both and the test's 90% lower bound above −2% with 50+ bets (test-only studies: 100+ bets and lower bound above 0); conditional = pays only if the venue prices that market like a naive model; reference = hindsight upper bound, not tradeable; weak = test profitable but uncertain; rejected = loses out of sample. ROI is profit per $1 staked. With many ideas tested, some "wins" are luck — only trust test results, and paper-trade promising ones before real money.

Forward tests (live paper)

Lab notebook

Loading…