How we count
Where the data comes from
Market data comes from Kalshi’s public API. We are not affiliated with Kalshi.
Official transcripts of the President’s remarks come from govinfo.gov, usually 4–7 weeks after the event: see Official transcripts.
Official transcripts spell some market words differently, so their counts on this site include those spellings: “health care” for Healthcare; “Dumocrat” for Dumbocrat / Dumacrat; “Zelenskyy” for Zelensky; “cease-fire” for Ceasefire; “anti-Semitism” for Antisemitism; “anti-Semitic” for Antisemitic; “McDonald’s” for McDonalds; “401(k)” for 401K. The list holds only spellings of the same spoken word. It changes the official counts only, not k/n, My transcripts or Transcribe.
When the White House corrects a slip in brackets, as in “Mandami [Mamdani]”, the official counts take both words, and the quotes show both.
Prices of events that start within the next 24 hours, or started less than 6 hours ago, update every 20 seconds. Everything else updates every 10 minutes: new events, new words, results and the other prices.
k/n: how often a word was said
k/n means: said in k of n past events of this speaker where this exact word was a market. A word counts as said when its market settled YES.
On ABC World News Tonight a word counts when any host or reporter says it, so its k/n is counted over past shows of the program, not over one speaker. The first shows, in July 2026, counted only David Muir; to leave them out, type “reporter” under Title contains in Choose events.
On an earnings call a word counts when anyone speaking for the company says it, the call operator included; analysts asking questions do not count. Each company is counted on its own, over its own past calls on Kalshi. Most companies have only one to four past calls, so most of their words carry the small n label. We have no transcripts of past calls: the results are Kalshi’s.
Words match only by their exact key: the same variants split by a slash, and the same “N+ times” threshold. So “Tariff” and “Tariffs” are two different words, and so are “Tariff” and “Tariff (3+ times)”.
Kalshi’s rules decide which forms of a word count: the word, its plural and its possessive, also inside a hyphenated compound, and the ordinal of a number in a phrase. For a word with slashes, any of its variants counts. We take each result as Kalshi settled it. How Kalshi settles the harder cases, and where our own counter differs from its terms, is on the Rules page.
k/n in the same format
The second k/n column counts only past events of the same format as this one:
- Rally
- Tele-rally
- Press conference
- Interview
- Meeting
- Speech / ceremony
We read the format from the event title and the market rules, before the event starts. Events with no clear format, ABC World News Tonight (every show is the same kind), earnings calls and windows have no second column.
In Last 5, a filled dot means the word was said and an empty ring that it was not; either one inside an outer ring is a past event of another format than this one.
Choosing events by hand
On the page of a single event, Choose events lets you pick which past events of the speaker to count: tick them, or type a part of their title. The Chosen column then shows k/n over those events only, by the same rules as the All column. The choice lives in the page address, so a link shares it; nothing is stored.
My transcripts on event pages
The transcripts you keep in My transcripts get columns of their own on the page of each single event of the same speaker: My transcripts, and My rallies or another format when the page has a column of its format. They count your transcripts dated on or before the day of the event (ET) that have Count on event pages on, never one saved from that event’s own Transcribe page. A transcript counts a word when it is said at least as many times as the market needs, by the rules of our counter. The columns are made in this browser and never change Kalshi’s k/n. With no transcripts of the speaker, the page has no such columns.
Transcribe
On the page of a single event, the Transcribe link opens a page that listens to the broadcast and marks the market words as they are said. After the event, the same page listens to its recording, played in another tab, and shows Kalshi’s results next to what it heard. Your browser sends the sound of a tab you share, or of your microphone, straight to Deepgram, a speech-to-text service, on your own Deepgram key. The key, the sound and the text stay between your browser and Deepgram and never reach our server.
To turn any recording of a speaker into one of your transcripts, press Transcribe a recording on My transcripts: play the recording, on YouTube or in any player, in another tab and share that tab, then choose the speaker, the date of the recording and its format, and save it. It counts in the My transcripts columns of that speaker’s events on and after its date, like any transcript you keep. A recording is transcribed as it plays, so an hour of recording takes an hour and is priced like a broadcast, by the minute played. The site never downloads or keeps the recording; a link you add is kept as text in this browser. For a video on YouTube there is a faster way: + Add transcript on My transcripts links to NoteGPT, a third-party site that makes a transcript from the video’s link; its text has no speaker names, so you keep only the speaker’s part yourself.
Deepgram charges your account about $0.46 an hour at its list price; new accounts get $200 of free credit. A word counts when it is heard in a form Kalshi’s terms count: the word, its plural or its possessive, also inside a hyphenated compound, and the ordinal of a number in a phrase; a word with “N+ times” needs N mentions. The event page then shows it as heard, in this browser only.
With Separate speakers on, Deepgram also numbers the voices it hears, for about $0.12 an hour more. It cannot tell who they are: under the text you mark each voice as the speaker the market’s rules name, or as Someone else, and the count follows your marks, earlier mentions included. Until you mark anyone, everyone counts. In a market about one person, once you mark that person, the voices you have not marked stop counting; in a market about a group, such as the hosts and reporters of a show or the people of a company on its call, a voice counts until you mark it Someone else. The numbers can be wrong: two people can share a number, one person can get two, and after a reconnect the numbers start again.
Speech recognition makes mistakes, and it cannot tell whether a clip or an ad is playing. Switch counting off while one plays (and, with Separate speakers off, while someone else speaks), dismiss false mentions, and always check the stream yourself.
Ask AI
Ask AI is an AI agent that answers questions from this site’s data: the rules and the case library, how often a word was said at past events (k/n, Last 5 and the past events behind them, without prices), the events of a day, the official transcripts and, if you let it, your My transcripts. It runs in your browser on your own OpenRouter key: your question and its answer go between your browser and OpenRouter, and our server only answers the lookups the model makes. It gives no prices, chances or trading advice. Your chats are kept in this browser only.
Which events count
Only events that were held count. Three kinds of events are left out:
- events where “Event does not qualify” settled YES;
- cancelled events;
- events where every word settled NO and there was no “Event does not qualify” market.
The event page shows them on their own line, as “Not held: x of y”.
No look-ahead
Each event is compared only with events that took place before it. The page of a past event shows k/n as it was before that event started.
Windows (“this week”, “this month”, companies in a month, nicknames before a date) are never mixed with single events. A window is compared only with past windows of the same kind: weeks with weeks, months with months, company windows with company windows, nickname windows with nickname windows.
Small samples
When n is below 5, the word gets a small n label. A few events say little. A word that no held past event of this speaker had as a market is marked new word.
Lowest prices
Click a word to see its last 5 events under it, newest first. For each one we show the lowest trade price in the 24 hours before the start, leaving out the hour when the market had just opened, and the lowest trade price during the event. Both come from hourly candles. Then the volume in contracts.
On the page of a settled event, Before start is the last trade price before the start: the close of the last hourly candle in the 24 hours before it, leaving out the hour when the market had just opened.
During an event, a word whose yes bid is 99¢ or more is struck through on a grey row under Said at the top of the table, as on Kalshi. If its bid then falls below 99¢, the word moves under Fell back, without the strike, with a red unbond label.
What we keep
To count visits, the first page you open sets a cookie, mid, with a random number that lasts a year. We keep only a hash of it with the day, the page, the event and the ref of the link you came by, and no log of addresses or searches.
Your keys, Ask AI chats, My transcripts, marks and settings, such as the theme and the time zone, are kept only in this browser and never reach our server. Settings shows them and can erase them all.
What this is not
Not financial advice. We show what happened in the past, not forecasts. There is no Edge and no buy or sell advice here.
Times are in ET, New York time, as on Kalshi, and in a second zone you pick in Settings: UTC until you do. The choice is kept in this browser.