What a temperature market actually settles on

Three things sit between a forecast being right and a temperature contract paying out, and none of them is the weather. There is a single sensor at a single named site. There is a day that does not start at midnight on your clock. And there is a text product, published by a specific organisation on a specific timetable, which the contract treats as the truth and which the organisation that publishes it describes as preliminary.

A reader who has followed the forecast closely and got the temperature right can still lose the contract on any of the three. This page is about the machinery, not about the weather, and the documents it is built from are the ones the exchanges filed: Kalshi's certified terms for its temperature rulebook, Polymarket US's rulebook and its own weather documentation, and the National Weather Service instruction that specifies the product both of them have pointed at.

How it works

The chain from a thermometer to a settled contract has five links, and every one of them is written down somewhere public.

1. An instrument reports. Most official US observations come from an Automated Surface Observing System station, generally at an airport, sampling continuously.
2. The station transmits a daily summary. The ASOS Daily Summary Message carries the day's maximum, minimum and other elements. NWS Instruction 10-1004 sets the primary transmission time at 00:15 a.m. local standard time for each ASOS site, and says plainly that the message "is not subject to WFO quality control" — WFO being the local Weather Forecast Office.
3. A forecast office publishes the Climatological Report (Daily), the CLI. The AWIPS Climate Program builds it from that summary message. The instruction requires at least two issuances a day: the first between 12:30 a.m. and 5:00 a.m. local time, covering the previous calendar day, and a second in the late afternoon covering the current one. Temperatures print in whole degrees Fahrenheit, with the time of occurrence in local standard time.
4. A venue reads that product at a time it specified in advance, and computes the contract's expiration value from it.
5. Months later, somebody else decides what actually happened. The National Centers for Environmental Information runs the quality control that makes the record final. Nothing about that stage reaches the contract.

The day is not the day

The climatological day runs midnight to midnight local standard time, and it does not move for daylight saving. The Chicago forecast office states the consequence in one sentence: "This does not change during daylight saving time, so the daily observation period during daylight saving time spans from 1 AM to 1 AM local daylight time."

For roughly eight months of the year, then, the window a US temperature contract measures begins and ends an hour later than the reader's clock says. A warm hour just before midnight on the final evening belongs to the following day's contract. Kalshi's certified terms carry the same convention and make it general: "All time periods are interpreted in the local standard time of <area> unless otherwise specified," with the end of the period defined as 11:59:59 PM local standard time on its final day.

The number is preliminary and says so

The instruction governing the product is explicit: "The data in these products are preliminary since the products are issued before all levels of NCEI QC makes the data final." The Chicago office's public FAQ puts a duration on it — observations do not become official until a final quality control process conducted by NCEI "3+ months after the observations are taken."

The correction policy for the daily report is section 4.1.4 of the instruction in its entirety, and it reads: "These will be done as needed." Corrections to the archive itself run through a ticketing system called Datzilla. Even the monthly form a forecast office marks as its "final posting" carries a caveat in the instruction that the phrase "does not mean the data are 'final,' as defined in section 1." In the daily station archive, GHCNd, real-time sources are typically replaced by archive-ready ones 45 to 60 days after the end of a month.

So the number a contract settles on is the first one published, by a program reading an unreviewed automated message, from a network whose custodian will not call the value official for a quarter.

Two venues, one question, two machines

Both venues below list a contract that a reader would describe the same way — the day's high in a named city. The settlement machinery is not the same, and the differences are the kind that decide a contract rather than colour it.

Polymarket US names the product, the station and the clock. Its weather documentation states that "Settlement is determined by the official NWS Daily Climate Report (CLI) published by the local Weather Forecast Office," and names one station per city: KNYC at Central Park for New York, KMDW at Midway for Chicago, and the airports — KSFO, KMIA, KLAX — for San Francisco, Miami and Los Angeles. Settlement occurs at 8:00 AM ET on the following day. If the CLI reading "is inconsistent with the 24-hour METAR observation for the same location, settlement may be delayed until 11:00 AM ET for review." If nothing publishes within a week, the contract settles at last fair market prices. Note the Chicago choice: the city's two reporting airports do not agree with each other, and the contract picks one.

Kalshi names the mechanism and keeps the source agency out of the public filing. The certified terms for its LOCALTEMPERATURE rulebook, self-certified to the CFTC on 5 March 2026, say the underlying is the temperature in the area "as reported by the applicable Source Agency", that the Source Agency "will be specified by the Exchange at market listing", that the exchange "may list multiple source agencies with a hierarchical ranking", and that where there is a hierarchy "the highest-ranking available source shall govern". Which agency that is sits in Appendix D, filed with a request for confidential treatment. The public part of the filing adds that "A new Source Agency can be added via a Part 40 amendment."

Four more clauses in that filing decide contracts:

• First publication governs, with a window for corrections. "Only the first official report or data release by the applicable Source Agency that includes the relevant data will be used for resolution, except where that data is retracted, corrected, or identified as erroneous by the Source Agency or the Exchange prior to the Expiration Date, in which case the Exchange will use the corrected data."
• After expiration, nothing counts. "Revisions, corrections, or quality control adjustments made after the Expiration Date and Expiration time shall not be considered, and the data as it appeared at the Expiration Date and Expiration time shall govern." The expiration time is 10:00 AM ET.
• Precision is the source's, not the headline's. "Contract resolution is based on the full precision reported by the Source Agency. Rounding by media outlets, secondary reporting, or third-party summaries does not affect resolution." The comparison operators are defined against that precision too: "exactly" means equal "when rounded to the precision reported by the Source Agency".
• If no source produces a number, the contract does not resolve on the weather at all. "If no data can be obtained from any Source Agency by the Expiration Date all strikes shall resolve at the last fair price as determined in the sole discretion of the Exchange."

Polymarket US's rulebook reaches the same destination by a different route: its definition of Source Agency is the exchange itself, referencing external sources designated in a contract's terms and arranged as primary, secondary and tertiary, with the exchange able to designate additional or substitute ones. Rule 10.4 lets named officers review any circumstance affecting the reliability of the underlying event and determine the final outcome. Neither venue's resolution is mechanical all the way down, and both say so in the document rather than in the interface. The general version of that argument is on who decides the outcome.

Contracts reached through a broker inherit the exchange's rules rather than the broker's: the weather contracts on Robinhood are routed to a designated contract market, and the rulebook that settles them is that exchange's.

What changed when the source agency changed

On 27 August 2026 The Weather Company announced a partnership in which Kalshi uses its "enterprise-grade weather data feeds as its trusted source for verifying weather-related market outcomes", and states that it "provides the authoritative observation data used to settle these markets with a consistent, documented methodology for each market type". The announcement runs the other way too: Kalshi's probability data appears inside The Weather Channel app and weather.com.

Describe that swap only in terms of what a reader can and cannot see, because that is the part that is checkable.

A government product has a published specification — the instruction cited throughout this page is a numbered, dated, public document that says what the CLI contains, when it is issued, what quality control it has and has not had, and how corrections are made. It is free to read at the moment the exchange reads it, and a reader who disagrees with a settlement can point at a text product and a directive paragraph.

A commercial feed is a contract between two companies. The methodology may be documented, as the announcement says it is, but it is documented to the customer. Pricing for that vendor's standard API package is published — $500 USD/month for 1 million calls on an annual plan — while its cleaned historical and history-on-demand products are listed as enterprise items arranged with its sales team. What a reader gets for free from the National Weather Service, they do not get for free here, and the one number they most want to check after a settlement is the one furthest inside the enterprise tier. Where weather data comes from works through what each source costs and what its licence allows.

None of this is a claim about why the change was made. It is a description of which documents a reader can hold in their hand afterwards.

What you can do about it

Read the series rules before you read the price. Four facts decide the contract and all four are published: the station, the source agency, the cutoff time, and the treatment of corrections. On Kalshi the general form is the rulebook's certified terms and the specifics are in the market listing; on Polymarket US both are in the weather documentation. If a page you are reading summarises a contract without naming a station, it is not describing the contract.

Check which clock the contract is on. If the period is local standard time — and on US temperature contracts it generally is — then between March and November your day and the contract's day are an hour apart. That matters most for the late-evening hours at the end of the period and for the overnight minimum.

Line the cutoff up against the publication timetable. The daily climate report's first mandatory issuance is between 12:30 a.m. and 5:00 a.m. local time. Kalshi's expiration time is 10:00 AM ET, Polymarket US settles at 8:00 AM ET with a review path to 11:00 AM ET. The gap is where a correction still counts — and the size of that gap differs between the two venues for the same calendar day.

Ask whether you can read the settlement number yourself. This is the practical consequence of a source agency that is a commercial feed rather than a public product. If the answer is no, you will learn the settlement value from the exchange rather than verify it against the exchange, and that is a different position to be in when you disagree.

Look at how often the preliminary number moves before you assume it does not. The station archives are free: GHCNd and the Integrated Surface Database both publish per-station history, and comparing the first-published daily value against the archived one for a station you trade is an afternoon's work that tells you whether the correction clause is theoretical or live for that site.

Treat a price gap between two venues as a rules question first. Two contracts that read the same in their titles can differ in station, cutoff and correction treatment, and any of the three moves a fair price without anybody being wrong about the weather. The general case is why the same contract costs two prices; the weather case is unusually easy to check, because both sets of rules are published.

Do not price a forecast against a contract until you know the distance between them. A model gives you a gridded value at fixed hours; the contract wants one station's maximum across a shifted 24-hour window at the precision the source reports. What an AI weather model gives you is about that gap specifically.

For the venues themselves and what each one's resolution field resolves to, the prediction market venues section carries the per-venue detail, each dated to the document it was read from.