Piece 02 · Kalshi weather · the trade tape

Who collects the cent on a dead contract.

Once a station's running high has passed a temperature bracket, that bracket's No is worth a dollar and its Yes is worth nothing. Kalshi's public trade tape shows 3.1 million contracts changing hands on such brackets in 47 days, almost all of them at a cent. Somebody is selling those cents. We measured who, how much a day, and whether anyone else could get a turn.

Published 2026-09-17 · Kalshi trades 2026-07-24 to 2026-09-08 · order books read 2026-09-16 18:47 UTC · method at the end

Dead brackets
1,938
in 47 days, 1,641 traded after the kill
Traded after the kill
3.13M
contracts, $46,442 at the Yes price
At one or two cents
2.99M
$30,002, on every one of the 47 days
The seller's net
$608
a day, about 1% a day on $55K of collateral
Already resting at 1c
43,843
contracts, the median dead bracket
Ahead of a new ask
32 days
of that bracket's flow; it settles in one

The previous piece found that a Polymarket wallet makes its living on dead brackets: the moment the day's running temperature rules a bracket out, it sells that bracket's Yes into whatever bids are still resting there. On Kalshi that trade prints at size about twice a month. But the tape on those same dead brackets is anything but quiet. It is busy every day, at a price nobody writes threads about.

A dead bracket

Kalshi's daily temperature ladders are brackets: 87–88°F, 89–90°F, at or above 91. Once the station's running high for the day has reached 90, the 87–88 bracket cannot settle Yes. We call it dead when the confirmed running extreme, the second-most-extreme reading of the day so that one bad reading cannot do it alone, has cleared the bracket by a full degree, which absorbs a whole-degree Celsius reading taken as exact. On the desk's 47-day ledger that rule marked 1,938 brackets. Two of them settled Yes anyway, both on readings three and four degrees off, so the rule is right 999 times in a thousand and wrong when the feed is.

What trades on it

From the first half-hourly read at which the rule fired to the market's close, Kalshi's tape shows 13,735 trades on those brackets, 3,125,236 contracts, $46,442 at the Yes price. The distribution is not a curve. It is a spike at one cent and a long, thin tail.

Contracts traded on dead brackets, by the Yes price printed

47 days · from the kill to the close · the 1–3c bar is 97% of all contracts
the numbers
source: Kalshi /markets/trades on 1,938 brackets ruled out by the observation with a 1°F margin

Who takes which side matters here. A taker who buys No at 99c has sold a dead bracket's Yes into a resting bid: the wallet's trade. A taker who buys Yes at 1c has paid a cent for a contract that will pay nothing. The tape splits cleanly: takers bought Yes on 13,225 trades and 3,012,579 contracts at an average 1.2c, $35,355 in all; takers bought No on 510 trades and 112,657 contracts at an average Yes price of 9.8c, $11,087.

Share of each price band that was Yes sold into a resting bid

the rest was Yes bought by a taker, paying for a dead contract
the numbers
source: the same tape; taker side from Kalshi's trade records

Above 3c the tape is mostly the wallet's trade and it is small: 80,655 contracts in 47 days, two days carrying nearly all of it. At 1c the tape is the opposite trade, three million contracts of it, and it is who buys a dead contract for a cent that explains the whole thing. They are not lottery players. They are people closing a short Yes: a market maker who sold that Yes at 20c earlier in the day, buying it back at 1c to release the 80c of collateral behind it, or to be rid of the small chance that the settlement disagrees with the observation. Somebody is on the other side of every one of those cents, resting an ask at 1c and getting filled.

The seller's arithmetic

Suppose one account had been every 1c and 2c ask on every dead bracket in those 47 days. It would have sold 2,989,079 contracts for $30,002. It would have lost $1,423 on the one wrong bracket that traded, Boston on August 30, where the running reading said 86 and the settlement said 83. Net $28,579, or $608 a day. Each contract sold at 1c needs 99c of collateral until the market settles, a median of 17 hours later, so a median day ties up about $55,000 to earn that $608: a little under 1% a day. And 80% of the flow prints more than an hour after the kill, so it does not need the wallet's speed. It needs the observation, the collateral, and a place in the queue.

The queue is the whole story. Kalshi fills resting orders by price and then by time, and at 1c on a dead bracket the line was already very long before we arrived.

The queue

On September 16 at 18:47 UTC we read the order book of every bracket the rule had ruled out at that moment, twenty-four of them across the ladders. Every one had no Yes bid at all and a Yes ask at 1c. Behind that ask, already resting:

Contracts resting at the 1c ask on each dead bracket

one pass, 2026-09-16 18:47 UTC · lows in green, highs in blue
the numbers
source: Kalshi order books; a No bid at 99c is a Yes ask at 1c; sizes in contracts

The lows carry 9,385 to 19,600 contracts each. The highs carry 40,593 to 210,822. The median is 43,843; the total is 1.5 million contracts, which is $1.5 million of collateral somebody has posted at a cent. Against it, the tape's entire flow of 1c buying was about 55,700 contracts a day across roughly 41 dead brackets, some 1,400 per bracket per day. A new ask at 1c stands behind a median 32 days of that flow, on a bracket that settles within one. It is never reached. There was no Yes bid on any of the twenty-four, so there is no taker route either.

So who collects it

Whoever posts six-figure asks at 1c the moment a bracket dies, which is the market makers, by right of being first. The $608 a day is theirs, and it is small money for the collateral it takes. The only other way to earn anything on a dead Kalshi bracket is the wallet's: a stale bid at 3c or better, which the tape shows printing at size twice in 47 days, hours after the fact, to whoever is watching every ladder at that moment. We now log every dead bracket's queue and every stale bid on it hourly, and the scorecard shows both week by week.

Method

Dead brackets: the desk's own observation feed (NWS hourly observations plus the METAR six-hour extremes), the confirmed running extreme, and a 1°F margin, applied to every locked read in the 47-day ledger; the first such read per bracket is the kill. Tape: GET /markets/trades for each bracket from the kill to the close, with price, size, taker side and time. The kill is a half-hourly read, so the true moment is up to thirty minutes earlier and trades in that gap are not counted, which makes every number here conservative. Fees: Kalshi's taker fee is included in "what the sellers cleared"; makers pay a fraction of a cent per contract at this price. Queue: one pass of the ladder logger's order-book reads on the brackets the same rule marked dead at that moment; the logger now records it every pass.

Code, with tests: agents/backfill_tape.py, agents/dead_brackets.py, agents/kalshi_ladder_logger.py, agents/exp_queue.py. The full research note is the long form.