The Seats

Every week the board lists a hundred-odd seats. A seat is one trading
rule, running on one kind of market, with its own clock and its own money.
This page says what each one actually does — the real trigger, the real
threshold — so you can check our claims instead of trusting them.

Nothing here moves real money. Every seat trades on paper at the price the
order book was actually showing, and pays the fee Kalshi would actually
charge: ceil(7 × price × (100 − price) / 10000) cents per contract. That
fee is why a seat can pick winners and still finish red, and it's the most
common thing this lab has proven so far.

Crypto 15-minute markets

These settle every fifteen minutes on whether Bitcoin is above a strike.
They're the lab's highest-frequency pond, polled four times a second.

Buy the favorite — buys any side priced 90¢ or higher with at least two
minutes left. This is a control, not a hypothesis. Heavy favorites win about
93% of the time, which sounds wonderful until you notice you're paying 90¢+
to win 100¢ and the fee eats the rest. It exists to measure exactly how much
the obvious trade loses.

Chase the jump — when Bitcoin moves $40 or more within a 15-second
window, buys the side that move favors, but only if the market hasn't
already repriced (ask between 15¢ and 85¢) and there are at least two
minutes left. The question: does the order book lag a real price move
enough to trade against?

Last-minute entry — in the final 5 to 55 seconds, if our projected
settlement clears the strike by at least $10, buys that side at up to 85¢.
Tests whether the crowd is still selling a near-certain outcome cheap.

Fade the move — with about four minutes left (a 225–255 second band),
buys the underdog at 45¢ or less when window volume is unusually high.
The hypothesis is that late volume spikes are panic, and panic overshoots.

Buy the extremes — the second generation of Fade the move, born after
the first showed promise in a narrow slice. Same four-minute band, but only
in the wildest ~10% of windows by volume, and with no price cap. Sharper
version of the same bet.

Retired seats: Buy the longshot (sides under 10¢) and a 10–19¢ value rule
were both cut in July 2026 — the longshot verdict was +0.49¢ per trade,
which is a statistical zero. Their histories are kept; we don't delete
losers.

Crypto daily

Same idea, longer clock — daily Bitcoin, Ethereum, and Solana strikes.

Favorite and longshot are again controls: buy ≥90¢ with 10+ minutes
left, or buy ≤5¢ tails with 20+ minutes left. Model is the actual
hypothesis: it buys when a model built from our own tick history disagrees
with the market by 8¢ or more after paying the ask. Implied vol seats
(BTC, ETH, SOL) trade when the option-implied volatility disagrees with what
the market is charging.

Whale seats

The largest group, and the most specific. When a single trade of $500 or
more prints, the lab papers both directions — following and fading —
tagged by market family and by the price the whale paid.

Bands: 3–34¢, 35–65¢, 66–85¢, 86–97¢.

The bands exist because of a finding: an early hourly-follow seat looked
like +5.5¢ per trade, but that average was blending a genuinely profitable
band near 27¢ with a disaster above 90¢ where the whale was steamrolling a
market that was already right. Averages hid it. Now every combination of
family, band, and direction is its own seat with its own clock, which is why
there are so many rows.

Families: crypto 15-min, crypto daily, esports, tennis (tour), tennis
(challenger), MLB, MLB totals and spreads, soccer, basketball, football.
Out-of-season sports sit pre-wired and arm themselves the day their markets
list, so no season opener is missed.

Challenger tennis carries the single largest whale tape on Kalshi, which is
why it produces both the best and worst seats most weeks.

Sports

Tennis favorites — buys the favorite at the ask when it's priced between
65¢ and 92¢ and the spread is 4¢ or tighter. Day-two calibration found
favorites in that range winning far more often than priced, and longshots
losing badly; this seat measures whether that survives fees.

Esports fallers — watches Valorant, League, CS2, and Dota. When a side
that was a 65¢+ favorite earlier in the match is now offered at 20–35¢, it
buys the fall. Requires the condition on two consecutive polls, so a single
bad tick can't trigger it. The other esports seats test clutch rounds,
momentum, favorites, and longshots in the same markets.

Fade the combo — Kalshi's parlay-style combination markets. Sells the
combination when the implied odds of all legs together look more expensive
than the legs individually.

Weather

Six cities, on daily high-temperature markets. Each seat compares the
market's price against a weather-model ensemble and enters when the
disagreement clears a threshold. This is the lab's most honest cautionary
tale: the weather desk has had strong single weeks while sitting deeply
negative all-time, which is precisely why every board shows both numbers.

Market making

Rather than betting on outcomes, these seats quote both sides and try to
earn the spread. They pay no taker fees, which is why the maker desk shows
zero in the fee column — and why it has quietly become one of the better
performers. Variants differ in how aggressively they quote, how they skew
around a fair value estimate, and how fast they pull quotes when the market
moves against them.

Cross-market

Polymarket gap — the same real-world event often trades on both Kalshi
and Polymarket. When the two disagree by 5¢ or more, this seat takes the
cheaper side, testing whether the gap closes before settlement.

What would prove a seat wrong

A seat earns its place by beating fees over a large enough sample, not by
winning often. Two of the highest win-rate seats on the board are among the
biggest losers — buying 90¢ favorites wins 93% of the time and still loses
money. When a seat's honest clock says it doesn't work, it gets benched and
the verdict gets published. That's the whole point of the lab.