Perpetuals.

Kalshi’s third exchange lists perpetual futures — gold, silver, and twenty-one crypto pairs — levered, never expiring, and charging a funding payment every eight hours for as long as a position lives. Pit is the cluster’s agent there. It collected every candle and every funding rate before it asked anything, then asked fifteen pre-registered questions of twenty markets and got the answer the arithmetic predicted. It is armed with a $500 bankroll behind the same gate as Bookie and Quant, and holds nothing. This is a stream by design and not yet by revenue.

pit · kalshi perpetual futures Armed · nothing to trade

$ pit trade

at each funding window, 04:15 · 12:15 · 20:15 UTC · armed 2026-09-26 · figures as of then

collected36,000 hourly candles and 4,500 funding rates, 25 markets, back to the first listing on 2026-06-03

fundingevery 8 hours, zero 82% of the time, a median of 1.5 basis points when it is not

the fee12 basis points a side to take — a round trip costs sixteen windows of typical funding

rules15 declared before any result was read, held to windows they never saw, across 20 markets · 300 rule-market trials

survive0

nearestnone thick enough to judge — every held-out window is still under thirty trades

funding capture−0.73% a trade after costs over the full sample — the carry does not pay the round trip

held-outno rule has thirty trades yet; the most is sixteen, and at three windows a day that is months

bankroll$500 on the perps side and no more — the key cannot move money in, and the loop refuses a larger balance

trades0

The new thing a perpetual has is funding: an event contract charges its fee once at the fill, a perpetual charges every eight hours in whichever direction the rate points, so the funding series is not a detail of the backtest but the fee model itself, and Pit collected it before it declared a rule. The rules are the obvious ones, written down first: take the side that receives funding after it has persisted, and Quant’s momentum and mean reversion on eight-hour bars, which inherit Quant’s 3,645 trials because the plan said they would. Basis — the perp against spot — is declared and not run, because the spot archive ends six months before the first perp listed and the plan allows no proxy. Every cost is overstated: the taker fee and half the spread on every change of position, funding at the rate the exchange charged. The demo exchange corrected the plan three times before the live key existed — an order cannot open an isolated position, a read-and-trade key cannot fund a subaccount, and a market order at the exact quote can fill nothing — so the bound is the balance itself, every opening fill gets a stop the exchange holds, and every order goes out inside a kill switch sized to its own count. Pit wakes at each funding window, reads its journal, finds nothing that survived, and exits before it reads a key.

Fig. 01 — Pit, which measured the fee before it asked the question.