Kalshi Source Agency 2026: 44% Are Blank
We read the Kalshi source agency clause in all 3,249 contract documents. For 43.9% of markets, the body consulted first is still an unfilled blank.
Affiliate disclosure: contains affiliate links; we may earn a commission at no extra cost to you. We have no commercial relationship with Kalshi and earn nothing from it. 18+ only — see BeGambleAware.org. Last checked November 2026.
Key takeaways
- For 7,328 markets — 43.9% — the Kalshi source agency consulted first is a blank. We pulled every contract document on the board and read its Source Agency clause. The highest-ranked authority, the one asked before any other, is an unfilled template token on nearly half the exchange.
- Sports is the worst affected and Mentions is the extreme. 2,873 of 4,283 Sports markets (67.1%) have a blank top-ranked source. In Mentions it is 472 of 502 — 94.0%. Economics is the cleanest at 4.7%.
- Ranking is the part nobody mentions. 901 documents list their agencies “in hierarchical order”, covering 6,245 markets. A lower-ranked outlet is only consulted if every one above it fails — so a blank at the top is not one missing name among several.
- 680 of the 3,221 live documents are copies of another file. Checksumming all of them gives 2,541 distinct bodies, a 21.1% alias rate. One 2,331-word body is published under 168 different filenames. A further 28 documents 404, including one with no filename at all behind 19 markets.
- The answer does exist — in the API, not the document. Of those 7,328 markets whose PDF leaves the top source blank, all 7,328 carry a named settlement source with a working URL on the series object. This is a documentation gap, not concealment.
On this page
Every guide to trading prediction markets gives the same instruction, and it is good advice: before you buy, open the rules and find the Kalshi source agency — the organisation whose published number decides whether your contract pays a dollar or nothing. Kalshi's own help centre uses the term, and says settlement waits on “the source agency” to publish finalised results.
The stakes are the entire position, because this is the one clause with no second opinion. If the named body reports a figure, that figure is the settlement, and revisions filed after expiration are explicitly ignored. Get the agency wrong and you have priced a different question from the one you bought.
So we checked whether the document actually tells you. We downloaded all 3,249 contract documents the exchange points at, extracted the text of the 3,221 that still exist, and read the Source Agency clause in every one. On 7,328 of 16,695 markets the first-ranked agency is not a name. It is a blank in angle brackets.
We have form on this field, and it cuts both ways: our own contract-terms census established two days ago that 90% of markets share their document, and the full checksum below revises its duplication figure upward. This article is the same board read one clause deeper.
What a Kalshi source agency actually is
Kalshi groups markets into series, and the legal text attaches at series level — the same level as the fee coefficient and the prohibited-trader list. Each series object carries a contract_terms_url, and that PDF contains a labelled field reading Source Agency: or Source Agencies:.
The clause does two things. It names who reports the outcome, and where more than one body is listed, it ranks them. The wording is consistent: “The Source Agencies are, in hierarchical order, …”, followed by a list, followed by a rule that a lower-ranked agency is consulted only if every higher-ranked one is unavailable, unresponsive, has not reported by expiration, or is judged incomplete at the exchange's discretion.
A filled example, from the single-market document for Rhode Island election markets, ranks eight bodies: the Rhode Island Board of Elections first, then the Associated Press, NBC News, CNN, ABC News, CBS News, Fox News and The New York Times. It also gives the URL to check. That is the clause working exactly as the guides describe it.
What nobody has counted is how often you get that version.
43.9% of markets: the deciding body is a blank
We classified the first-ranked entry in each clause, because under a hierarchy that is the one that decides. A clause can name four backup newsrooms and still leave the primary authority as a parameter — SPORTSCHALLENGE.pdf ranks <person> first, then that person's social accounts. Counting any named outlet anywhere in the sentence would score that document as filled, which is why we did not.
| Highest-ranked source agency | Documents | % of docs | Markets | % of markets |
|---|---|---|---|---|
| Names a specific body | 2,022 | 62.8% | 8,190 | 49.1% |
| Names a role, not a body | 241 | 7.5% | 969 | 5.8% |
| Unfilled template token | 942 | 29.2% | 7,328 | 43.9% |
| No Source Agency field at all | 16 | 0.5% | 208 | 1.2% |
The document column and the market column disagree sharply, and the reason is the sharing structure. Blanks cluster in the documents that cover the most markets: among the 22 files serving 100 series or more, half leave the top source unfilled, against 24.3% of single-market files. A template written for 200 markets cannot name one authority, so it does not.
The “role, not a body” row is a middle case worth separating out. ACHIEVEMENTHOLDER.pdf says the source is “the official governing body”; AIRLINEZYN.pdf says “official airline announcements”. Nothing is bracketed, so a checker looking for template tokens scores these as filled, but you still cannot name the organisation from the page.
Which Kalshi markets are worst affected
The blanks are not spread evenly. Split by the exchange's own category labels, the spread runs from under one market in twenty to more than nine in ten.
| Category | Markets | Top source blank | Share |
|---|---|---|---|
| Mentions | 502 | 472 | 94.0% |
| Sports | 4,283 | 2,873 | 67.1% |
| Commodities | 120 | 71 | 59.2% |
| Financials | 1,328 | 669 | 50.4% |
| Elections | 2,617 | 1,167 | 44.6% |
| Companies | 159 | 67 | 42.1% |
| Entertainment | 2,653 | 920 | 34.7% |
| Science and Technology | 397 | 125 | 31.5% |
| Politics | 2,545 | 776 | 30.5% |
| World | 137 | 14 | 10.2% |
| Climate and Weather | 535 | 54 | 10.1% |
| Crypto | 306 | 29 | 9.5% |
| Economics | 919 | 43 | 4.7% |
The pattern is not random, and it is not really about effort. Economics and Crypto settle on a short list of institutions that never change — the Bureau of Labor Statistics, CF Benchmarks, Pyth — so the template can hard-code them. Sports settles on whichever league governs a fixture that will not exist until the contract is listed, so the template leaves a slot. Mentions, at 94.0%, is the category where the exact word matters most and the document says least.
The hierarchy matters more than the list
This is the part every competing guide compresses to “the named source decides”. It is more structured than that. 901 documents, covering 6,245 markets, rank their agencies “in hierarchical order” — 28.0% of documents and 37.4% of markets.
Ranking changes what a blank costs you. In a flat list of five newsrooms, a missing name is one option among several. In a ranked list, the top entry is consulted first and the rest are fallbacks that only activate when it fails. So when ACQANNOUNCE.pdf ranks <company 1>/<company 2>'s own investor-relations site first across 26 series, the document is telling you the primary authority is a company it declines to name.
The discretion clause is real and it is in the template. A lower-ranked agency is consulted if a higher one is “unavailable, unresponsive, has not reported… or if the Exchange, in its sole discretion, determines that a higher-ranked agency's reporting is incomplete, erroneous, or otherwise insufficient”. Kalshi also reserves the right to replace a source whose reliability is materially affected, announcing the change on its site. Both clauses are standard in the filed documents and in the KalshiEX rulebook.
A further 23 documents, behind 92 markets, refer the reader to a confidential appendix for the source agency. That is a lawful filing practice rather than a defect, but the practical effect is the same: the name is not on the page you were sent to.
One document, 168 different names
Our 20 November census sampled 30 documents and found 26 distinct bodies, a 13% duplication rate, and said plainly that the sample was weighted and not a board-wide estimate. We have now checksummed all of them, and the real rate is higher.
The 3,221 live documents contain 2,541 distinct bodies. 680 URLs — 21.1% — are byte-identical copies of another file under a different name. 143 bodies are published under more than one filename, between them covering 823 URLs and 6,929 markets.
| Filenames sharing one body | Markets | Words | Served as, among others |
|---|---|---|---|
| 168 | 393 | 2,331 | ENTITYOUTCOME, NBAALLSTAR, CFBAWARDS |
| 71 | 288 | 1,232 | ESPORTS, TENNIS, BASEBALL |
| 46 | 290 | 889 | RANKLIST, GOLDENGLOBESNOM |
| 45 | 211 | 1,433 | SOCCER, NCAAB, NCAAF |
| 40 | 107 | 1,067 | SOCCERTOPX, UCL |
You can check the first row in two clicks. Open the college-football awards document and the title at the top of page one reads ENTITYOUTCOME — the generic file, served under a sport-specific name, to 167 other names as well.
This matters for a mundane reason. The filename is the only clue most readers have that a document might be generic, so a sport-specific name on a generic body removes the one signal that should make you check the fields instead. It also means the true coverage figure is worse than the one we published two days ago: 16,775 markets rest on 2,541 real documents, a mean of 6.6 markets per document, not 5.2.
Twenty-eight documents that are not there
Of the 3,249 URLs on the board, 28 return HTTP 404. We re-checked all 28 and every one was still dead. Behind them sit 80 live markets whose contract terms cannot be read at all: 38 in Sports, 21 in Politics, and the rest scattered across Economics, Entertainment, Science and Technology, Financials, Climate and Weather and Companies.
The largest single case is the most obviously accidental. Nineteen series point at assets.kalshi.com/contract_terms/.pdf — the directory and the extension, with no filename between them. The markets behind it are not obscure: Treasury bill rate, Japan Tariff Size, Central Bank Digital Currency, NFL hire coach. Three more dead files are named for what they are: FOOTBALLARCHIVED (15 markets), BASKETBALLARCHIVED (13) and OLDCRICKETSERIES (6), which look like archived templates that were removed while series still referenced them.
80 markets out of 16,775 is half of one per cent, and none of it affects what binds you — the product certification and the rulebook do that. But a required field that resolves to nothing is the one defect a trader cannot work around by reading more carefully, and it is worth knowing it exists before you assume you have mis-typed a URL.
Where the answer actually is
Here is the correction to the alarming reading of everything above, and we went looking for it before writing a word of the rest. The source agency is published for essentially every market. It is just not in the PDF.
The same series object that carries contract_terms_url also carries a settlement_sources array, and that field is populated on 16,768 of 16,775 series. Each entry has a name and a URL. Take the 7,328 markets whose document leaves the top-ranked agency blank, and look at what the field says for exactly those markets: 7,328 of 7,328 carry a working http URL, and 7,323 name something. Not one of them contains a template token.
| Series | Document says | Field says |
|---|---|---|
KXNFL1Q | “the governing league or association of <…>” | the Governing League → nfl.com |
KXGOVNHNOMD | “<political party> of <state>” | Republican Party → gop.com |
KXMADRID2ND | “the official election authority…” | Junta Electoral Provincial de Madrid |
KXBRASILEIROFTTS | “the league governing <game>” | ESPN → espn.com |
Two honest caveats. The field's name can itself be a role — “the Governing League” is not an organisation — but it always ships with a resolvable link, which is the part you can act on. And a URL you can open is not the same as a figure you can pre-check; we measured how far that gets you in our cross-venue settlement census, where 525 series name nothing but Kalshi's own homepage.
So the shape of the problem is a documentation gap, not a hidden rule. Kalshi publishes more per market than its competitors do. What it does not publish is any single artefact that reads like your market's rule — which is exactly the thing every guide, OddsShopper's included, tells you to go and open.
Checking your own market in one request
You do not need a key or an account. One GET to the public series endpoint returns the whole board as complete objects if you leave the category parameter off, which is how every figure above was produced.
curl -s https://api.elections.kalshi.com/trade-api/v2/series \
| python3 -c "
import json,sys
want='KXNFL1Q' # your series ticker
for s in json.load(sys.stdin)['series']:
if s['ticker']==want:
print('doc :', s['contract_terms_url'])
for e in s['settlement_sources']:
print('src :', e.get('name'), '->', e.get('url'))
"
Read the src lines, not the document, for who decides. Open the document for mechanics — what counts as the underlying, how ties and postponements resolve, when expiration is measured. If the document's Source Agency clause has angle brackets in it, that is not an error on your end and there is no better copy to find.
One mechanical note, because it cost us time: curl -sI on that endpoint returns 404 while a plain GET returns 200. The endpoint does not support HEAD, so a link checker will tell you it is broken when it is not.
My verdict
The advice to read the rules before you trade is right, and the place it sends you is wrong for nearly half the exchange. A field documented as the terms of your contract leaves the first-consulted deciding authority blank on 7,328 markets, reaches 94.0% blank in Mentions and 67.1% in Sports, is a byte-identical copy of another file 680 times out of 3,221, and 28 times does not exist.
None of that is concealment, and it would be unfair to write it up as such. The deciding body is named, with a link, on 16,768 of 16,775 series, in a field two lines away from the one everybody quotes. The failure is that the human-readable artefact and the machine-readable one have drifted apart, and only one of them is what readers are told to open.
Trade off the fields. Use the PDF for mechanics, take the source agency from settlement_sources, and if you are in a Sports or Mentions market, assume the document is describing a family rather than your contract until you have checked the link yourself.
Know who settles it before you take the position
Settlement sources, prohibitions and fees are published per market on both venues. Reading them costs two minutes; finding out afterwards costs the position.
Visit Polymarket.com →Affiliate link to Polymarket; we have no commercial relationship with Kalshi and earn nothing from it. 18+ only. Prediction markets are not available everywhere — check your own jurisdiction first.
Related guides
- Kalshi Contract Terms 2026 — the sharing census this article revises, from 3,245 URLs to 2,541 real documents.
- Kalshi vs Polymarket Settlement 2026 — the
settlement_sourcesfield read across both venues. - Kalshi Prohibited Traders 2026 — the 452-clause census of the same series objects.
- Kalshi Fees by Market 2026 — the per-series fee coefficient, read board-wide.
- Kalshi Payout Time 2026 — how long settlement takes once the agency reports.
- Polymarket Resolution: UMA Disputes — what happens when the other venue's oracle is challenged.
Kalshi source agency FAQ
What is a source agency on Kalshi?
It is the organisation whose published result decides the market. Kalshi's contract documents carry a labelled Source Agency field, and the help centre says settlement waits for the source agency to publish finalised results. Revisions made after expiration are explicitly not counted.
Where do I find the source agency for my market?
Take it from the settlement_sources array on the market's series object in the public API, not from the PDF. The field is populated on 16,768 of the 16,775 series we read, and every entry carries a name and a URL. The PDF is the right place for mechanics and the wrong place for this.
Why does the Source Agency line in my Kalshi PDF have angle brackets in it?
Those are unfilled template merge tokens, and they are common rather than exceptional. On 942 of the 3,221 live documents — covering 7,328 markets, 43.9% of the board — the highest-ranked source agency is a token rather than a name. It is not an error in your reader.
What does “in hierarchical order” mean for settlement?
It means the agencies are ranked, not interchangeable. The first is consulted first; a lower-ranked one is used only if every higher-ranked one is unavailable, unresponsive, has not reported by expiration, or is judged insufficient at the exchange's discretion. 901 documents, covering 6,245 markets, use that wording.
Can Kalshi change the source agency after I buy?
Yes, within limits set in the filings. The documents and the KalshiEX rulebook reserve the right to replace a source agency whose reliability or transparency is materially affected, with the change announced on Kalshi's website, and a new agency can be added by a CFTC Part 40 amendment. This is one reason to re-read the rules before adding to a position.
Which Kalshi categories have the most blank source agencies?
Mentions is the extreme at 94.0% — 472 of 502 markets — followed by Sports at 67.1% of 4,283 markets and Commodities at 59.2%. Economics is cleanest at 4.7%, then Crypto at 9.5% and Climate and Weather at 10.1%, because those settle on a stable list of named institutions a template can hard-code.
Does every Kalshi market have a contract terms document that works?
No. 28 of the 3,249 URLs on the board return HTTP 404, re-checked the same day, leaving 80 live markets with no readable contract terms. Nineteen of them point at assets.kalshi.com/contract_terms/.pdf — a path with no filename in it.
Is the same Kalshi document published under more than one name?
Often. The 3,221 live documents contain only 2,541 distinct bodies, so 21.1% of the URLs are copies. One 2,331-word body is served under 168 filenames: open CFBAWARDS.pdf and the title on page one reads ENTITYOUTCOME.
Method and sources
Every figure comes from one pass of Kalshi's public series endpoint on 22 November 2026, which returned all 16,775 series as complete objects (21.3 MB, HTTP 200, no key) with the category parameter omitted. We then fetched all 3,249 distinct contract_terms_url values, extracted text with pdftotext -layout, checksummed the normalised text, and parsed the Source Agency field by anchoring on the labelled line. Percentages described as “of markets” are over the 16,695 series whose document resolves.
Three caveats we will not paper over. First, we classify the first-ranked entry only; many clauses that we score as blank do name backup outlets lower down, and our figure is deliberately about the authority consulted first. Second, our “role, not a body” bucket is the noisiest of the four, because a clause can name real organisations in a parenthetical after a generic lead; we kept it as a separate row rather than folding it into either side. Third, kalshi.com's own web pages returned HTTP 429 to this environment throughout, as on 18 and 20 November, so the rulebook is cited from the CFTC filing and the help centre from its canonical article URL.
- Kalshi — Get series list: the definitions of
contract_terms_urlandsettlement_sources. - Kalshi help centre — Market FAQs: Kalshi's own use of “source agency”, and what happens when its data is delayed or revised.
- Kalshi — CFBAWARDS contract terms (PDF): the file titled
ENTITYOUTCOME, one of 168 names for one body. - Kalshi — Mention contract terms (PDF): a 206-market document whose Source Agency clause is a bare token.
- KalshiEX LLC Rulebook (CFTC filing, PDF): the rulebook layer, including the exchange's discretion over sources.
- OddsShopper — Kalshi resolution criteria: the “the terms name one specific data source” framing this article measures.
- DeFi Rate — How contracts settle: reported cases of a mis-entered resolution source being corrected after the fact.