State of the Lab — week ending Aug 9, 2026

This week the lab ran 123 seats across 5 desks and settled 27,819 paper trades. Best seat: Market-making · crypto 15-min, multi-asset. Roughest seat: Fade tennis (challenger) whales @ 86–97¢.

Last week this page ended with a question: can the lab lose by less than the toll? One week later, here is the answer. Across 28,000 settled paper trades and 123 seats, before fees, the book came out at +$3.16. Within a few dollars of dead flat. After fees it lost $405.

Sit with that, because it is the whole finding. A machine made roughly twenty eight thousand bets at prices the exchange was actually showing, and on the raw question of what would happen next it fought the market to a draw. The win rate across every seat was 54.0%. Then it paid $409 for the privilege of asking. Last week four fifths of the loss was fees. This week it is effectively all of it. The predictions stopped losing. The toll did not move an inch.

The one desk in the black did it again, and by more. Market making finished up about $96 across 6,771 fills, its best week yet, paying almost nothing in fees, because a resting quote that gets traded against is free on nearly every market this lab touches. The desk grew this week, and the most instructive thing on the whole board is what happened inside it.

Three of the new books run the same idea, configured three ways, side by side on the same markets in the same week. They appear on the board as book A, book B and book C. Book A made +$23.13 winning 85% of its trades. Book C made +$24.54 winning 85%. Book B, the same idea with different settings, lost −$16.91 winning 79%. Two green, one red, one idea. If the lab ran only book A I would be tempted to call it edge. If it ran only book B I would have benched the idea by Wednesday. Neither reaction would have been justified yet, and the only reason I know that is that all three ran at once. One good week from one book is not a finding. It is a coin that landed well.

The usual caution applies to every maker number on this page, and it is spelled out under the board: these are simulated fills, a simulation cannot model being picked off by someone better informed, and where this lab has run the same idea live the real result has come in below the paper one. Book B is what that warning looks like when it lands.

The rest of the board tells last week's story again. The whale grid lost $417, of which $348 was fees; its actual trading loss was about $69. Esports paid $27 in fees and lost $26, which means its predictions came out almost exactly even. The weather desk, back from the outage described last issue, traded 834 times and lost $30. Nearly every directional book on the board is now flat or close to it at the prediction and red at the till.

The toll itself got a permanent page this week: which markets charge makers, which do not, the rounding trap at small size, and how this lab briefly got its own fee arithmetic wrong. It is called Does Kalshi charge maker fees? and it exists because the fee line above is the single most important number the lab produces.

None of this is advice, and none of it moved real money. It is a machine placing paper bets at real prices, paying the fees it would actually have paid, published in full so you can check it. Last week the question was whether the lab could lose by less than the toll. It can. It can apparently stop losing entirely, everywhere except the till. So the question for next week is harder: can anything on this board clear the toll and keep the change.

Each desk name links to The Seats, which gives every strategy's real trigger and threshold plus the Kalshi tickers it trades, enough to run any of them yourself. The one exception is market making: which markets it quotes is listed, how it quotes them is not, and the About page explains why.

All 133 seats behind these five lines, with win rates, fee drag, all-time results and the per-trade CSV, are in the members edition. 8 seats retired to date; their records stay on the board.

The market-making seats are simulated fills, and a simulation cannot model adverse selection: the tendency of a resting order to be taken exactly when someone better informed wants the other side. Where this lab has run the same strategy with real money, the live result has come in materially below the paper one. Read the maker desk as an upper bound on what the idea could earn, not as what it would have earned.

Fill verification: 1,387 of 33,800 trades (4%) were checked against real order-book depth.

The rest are entered at the ask the exchange reported, but the lab does not record the book for every market. Challenger tennis, MLB, soccer, basketball and football have no depth archive, so those fills cannot be re-checked after the fact. Trades whose price level had no size behind it are discarded, not counted.

Every trade above was logged live, then sealed: each night the lab hashes its raw market archive and trade ledgers into an append-only chain. Rewriting any past trade would change every head after it, so check this one against last week's issue.

Chain head as of 2026-08-09T04:14Z: 76717f2f3be9c20f287d60076a9de5a6f467d1261455954633d0412c47581a14