Notes on trading Kalshi, the CFTC-regulated event-contract exchange. Written up from a multi-agent sweep of public sources in September 2026: Kalshi’s own API, documentation, rulebook and fee filings; academic work on prediction-market pricing; and first-hand accounts from people who trade it.
Everything here is research, not advice. The best-evidenced number in the whole survey is that takers lose about 1.12% of notional as a group.
Pages
- exchange-mechanics — the fee formula, position and rate limits, sharding, settlement quirks, and the incentive programs. The constraints any edge has to clear before it is an edge.
- market-families — ten recurring market groupings a trader could specialize in, with real series tickers, cadence, settlement source and counterparty.
- strategy-families — eleven approaches people have actually run, each put to two reviewers instructed to refute it. All eleven were refuted, and the refutations are more useful than the claims.
- rfq-access — a correction. The widely repeated claim that Kalshi’s request-for-quote surface is invisible to retail is false, and this is what a normal account actually sees.
The short version
On Kalshi you get paid for providing liquidity and charged for consuming it, and essentially nothing else survived testing. Fee incidence, not forecasting, sets the sign of your return. The measured maker premium accrues to firms holding rebate agreements who sit in front of the retail queue, and the books thin enough to still carry an edge are too thin to hold size.
Capacity, not edge, is the binding constraint nearly everywhere. Median resting depth at the touch is about four dollars of notional, and 84.8% of markets have no volume in a day.
Confidence
Not uniform, and the pages say so where it matters.
- Exchange mechanics are high confidence, from the published fee schedule, rulebook, API documentation and live series metadata.
- Market shape is high confidence for names and cadence, medium for volumes, which are a single snapshot and move.
- Anything attached to a named trader is low confidence. No dollar figure attributed to an individual in this research is auditable.
Numbers are as of 2026-09-07. Fee types, incentive programs, shard assignments and category availability all change, and several changed during 2026 alone.