A bot for Kalshi's NYC high temperature market
A Windmill bot for Kalshi's NYC high temperature market (KXHIGHNY) is an AI agent that runs your written rule on the schedule you set: it reads the National Weather Service's forecast and Central Park's observations and the day's brackets, buys one and holds it to settlement. Counted on October 22, 2026, our agents had 27 settled paper bets on this market, up $3.38 before fees and most of that from a rule that read no weather at all, so test yours the same way: free on paper, no card.
Kalshi lists a market on each day's high temperature in Central Park. Write your rule for it in a sentence or a paragraph, and a Windmill agent runs it on paper, unattended, against Kalshi's live order book and fees. Six rules of ours traded it from late August 2026. The two that bought the National Weather Service forecast's bracket won 5 of 15 bets; a control that bought the crowd's favourite and read no weather won 5 of 6. Every number is below, with each rule in its own words.
What the market asks
Kalshi's NYC series, KXHIGHNY, lists one event a day: a ladder of brackets for that day's high in Central Park. When we read Kalshi's API on October 22, 2026, the ladder for that day had six markets: 63–64°, 65–66°, 67–68° and 69–70°, and an open-ended one at each end, 62° or below and 71° or above. Exactly one of them resolves Yes. This is the rule of the top one, word for word:
“If the maximum temperature recorded at New York City (CLINYC) for Oct 22, 2026, is greater than 70° fahrenheit according to The Weather Company, then the market resolves to Yes.”
CLINYC is the National Weather Service's climate-report code for Central Park, and the series names The Weather Company as its settlement source. Kalshi adds, in the same rules: “Preliminary Weather Company data may be subject to rounding and conversion differences from the final reported value.”
Trading on a day's high closes at 1 AM Eastern the night after, and the market settles the next morning. A YES or NO contract pays $1 if it is right and nothing if it is wrong, and buying one pays Kalshi's fee, which the fee calculator works out for any price.
What an agent reads for it
An agent runs on a schedule you set, every 10 minutes at the fastest. For this market our agents read the National Weather Service's forecast for Central Park from api.weather.gov, or the latest observation at the Central Park station (KNYC), or a web search for both; then they read the day's ladder on Kalshi, every bracket with its bid and ask. One of them noted in its rule that Central Park's readings often post late. What agents can read lists every tool an agent has.
The rules our agents ran here
Each of these ran on paper from late August 2026, each order for 5 contracts, and each was paused by September 6. The quotes are the parts of each rule that apply to New York, word for word; the agent's page has the whole rule.
city-highs-nws: the forecast for today
“Trade TODAY's market only. […] Coordinates: New York (Central Park) 40.7789,-73.9692 […] Buy the bracket that CONTAINS the NWS forecast high, 5 YES contracts at the ask, only if that ask is at most 60 and you do not already hold a bracket in that event. Skip a city if the market closes in under 2 hours.”
weather-3city: the forecast for tomorrow
“New York (series KXHIGHNY, station KNYC, gridpoint OKX/34,45) […] Buy the bracket that CONTAINS the forecast high, 5 contracts, limit 3 cents above its ask, only if that ask is 45 or lower. If the forecast sits within 1 degree of a bracket edge, buy nothing for that city. Never trade today's market.”
city-highs-crowd-control: the control
“CONTROL for the weather family: same four markets, NO outside information. […] Buy the HIGHEST-priced bracket (the crowd's favorite), 5 YES contracts at the ask, only if that ask is between 45 and 65 and you hold nothing in that event. […] No web tools, no forecasts, no reasoning about weather.”
nyc-high-temp-websearch: web search and 10 points of edge
“(1) web_search the current NWS forecast and the observed high so far today at Central Park (KNYC), with the time of the reading. […] A bracket BELOW a high already observed today is dead — never buy it, and if you hold it, sell. (3) Buy a bracket only when the forecast (or the observed running high, once it is afternoon in New York) makes it clearly more likely than its ask price implies, with at least 10 points of edge; pay at most the ask, 5 contracts. (4) Take profit: sell a held bracket when its bid is above 85 or your thesis has broken.”
ewr-vs-nyc-highs: Newark against Central Park
“Newark Airport (series KXHIGHTEWR) and Central Park (KXHIGHNY) daily highs are settled 10 miles apart and almost always land within 2°F of each other, but they are traded by different crowds. […] buy 5 YES on the bracket whose ask is LOWER, if that ask is at most 55 and at least 10 cents cheaper than the other city's equivalent bracket; otherwise buy nothing.”
nws-dead-brackets: the high already observed
“Buy 5 NO on every bracket whose top is below the running high (they are dead) if the NO ask is at most 96; and after 21:30Z, buy 5 YES on the bracket that CONTAINS the running high if its ask is at most 80.”
To start from one of these, open its agent's page and copy it: the forecast rule or the control. Each trades four cities as written; tell the builder to keep only New York. It is a starting point, not a recommendation.
The record
Counted on October 22, 2026, our agents had 27 settled paper bets on KXHIGHNY, from August 29 to October 20, 2026, and 12 of them won.
| Agent | Settled on (2026) | Bets | Won | P&L before fees |
|---|---|---|---|---|
| weather-3city | Aug 31 – Sep 8 | 6 | 1 | −$3.45 |
| nyc-high-temp-websearch | Aug 29 | 1 | 0 | −$1.85 |
| city-highs-nws | Aug 29 – Sep 7 | 9 | 4 | +$1.18 |
| city-highs-crowd-control | Aug 29 – Sep 7 | 6 | 5 | +$7.90 |
| nws-dead-brackets | Sep 2 | 1 | 0 | −$0.05 |
| ewr-vs-nyc-highs | Aug 31 – Sep 4 | 2 | 0 | −$2.05 |
| another agent | Sep 3 | 1 | 1 | +$1.00 |
| highs-leader-afternoon-east (running now) | Oct 20 | 1 | 1 | +$0.70 |
| All agents | Aug 29 – Oct 20 | 27 | 12 | +$3.38 |
Together the bets cost $56.62 and paid $3.38 more than that, +6%, before Kalshi's fees.
The fees are booked to each agent, not to each city: on these bets they ran about 8–9¢ a 5-contract bet, Kalshi's fee at 40–60¢ prices. Each agent's page carries its figure after fees, across every city it traded; on October 22, 2026 they were weather-3city −$17.69, nyc-high-temp-websearch −$3.41, city-highs-nws −$15.28, city-highs-crowd-control −$4.42, nws-dead-brackets −$1.27 and ewr-vs-nyc-highs +$0.08. The control's +$7.90 before fees on 6 New York bets sits inside its −$4.42 after fees across four cities.
Why each rule did what it did
The control made the series' gain. city-highs-crowd-control won 5 of its 6 bets for +$7.90, more than all 27 bets together; without it the rest were −$4.52. It paid 57¢ a contract on average, and a favourite priced at 57¢ wins 5 of 6 about one time in five by chance alone. Across Chicago, Miami and Austin the same control lost money, $1.90 over its 30 bets in the four cities.
The forecast for today came out about even. city-highs-nws paid 42¢ a contract on average and won 4 of 9, close to what its prices implied: +$1.18 before fees here. Across its four cities its page shows −$15.28 after fees.
The forecast for tomorrow lost. weather-3city bought only when the forecast's bracket cost 45¢ or less, so it bought when the market disagreed with the forecast a day ahead. It won 1 of 6, −$3.45.
The rest barely traded. nyc-high-temp-websearch ran 121 times and placed 13 orders; one position went to settlement, a $1.85 loss. ewr-vs-nyc-highs's two Central Park bets lost $2.05 (its one Newark bet won $2.35), and nws-dead-brackets had one settled position here, a 5¢ loss. One bet by another agent, in September, won $1.00.
Running now. Two agents of ours have run on this market on paper since October 17, 2026. One buys today's leading bracket at 80–89¢ around 3 PM; it bought New York once, at 86¢, and won $0.70 on October 20. The other buys tomorrow's open-ended bracket when it is the favourite at 60–92¢, and as of October 22, 2026 it had placed nothing: “no open-ended tail favourite in the 60–92¢ YES-ask band”, in its own words.
One market, worked through
The high in Central Park on September 1, 2026, settled at 82°: the market “82° or below” resolved Yes, and Kalshi settled it at 7:19 AM Eastern on September 2. Three of our agents held a position in it:
- city-highs-nws read that day's forecast and bought “82° or below”, 5 contracts at 53¢, $2.65. It was paid $5.00: +$2.35 before fees.
- city-highs-crowd-control, the control, bought the same bracket as the crowd's favourite at 64¢, $3.20, and was paid $5.00: +$1.80.
- weather-3city, which trades a day ahead, held 83–84° at 32¢ and 85–86° at 14¢, $2.30 in all, and lost both. Its rule says to skip a city it already holds; it held two brackets in this event anyway.
A one-degree miss is a lost bet: the brackets are two degrees wide, and a forecast a day out is often further off than that. These figures are paper, counted on October 22, 2026.
Questions
- Is a bot for the NYC high temperature market profitable?
- We can't tell you that one is. Counted on October 22, 2026, our agents' 27 settled paper bets on this market were up $3.38 before fees, and the rule that made the most read no weather at all: 6 bets, too few to tell from luck, and across its four cities that rule's page showed −$4.42 after fees. Across New York, Chicago, Miami and Austin the same rules lost money. Running your own rule on paper is how you find out, before you risk money.
- Which station settles the NYC market?
- Central Park. When we read Kalshi's API on October 22, 2026, each market's rule named New York City (CLINYC), the National Weather Service's climate-report code for Central Park, and The Weather Company as the source of the reading. Not LaGuardia, JFK or Newark; Newark has its own series, KXHIGHTEWR.
- When does it close and settle?
- Trading on a day's high closes at 1 AM Eastern the night after. The market in the worked example above, for September 1, 2026, settled at 7:19 AM Eastern on September 2. The high itself is usually in by mid-afternoon.
- Can an agent read Central Park's observations?
- Yes. An agent can fetch any public https URL, so it can read the station's latest observation (KNYC) and the forecast for Central Park from api.weather.gov, and it can search the web. Observations come in degrees Celsius, and Central Park's sometimes post late. What agents can read lists every tool.
- What are the fees?
- Paper orders pay Kalshi's own fees, worked out the way Kalshi works them out. When we read Kalshi's API on October 22, 2026, the New York, Chicago, Miami and Austin daily-high series had fee type quadratic with multiplier 1, the same settings as Kalshi's 15-minute bitcoin and ether series: an order that takes a resting price pays the most at 50¢, 1.75¢ a contract, and less toward either end. On our agents' bets here the fees ran about 8–9¢ a 5-contract bet. The fee calculator gives the fee and the breakeven for any price.
- Paper or live?
- Agents trade on paper unless you build them live. Paper orders fill against Kalshi's live order book, pay Kalshi's fees and settle when the real market settles. To trade real money, connect your own Kalshi API key and build a live agent; your money stays on your Kalshi account.