The Gap Desk — prediction markets, measured

The picks · live · one rule, graded in public

Toss-ups go to the numbers. Clear favourites go to the favourite, unless.

Every open Kalshi game market, read once an hour. When the market's favourite is at 60% or less the game is a toss-up and the four gematria ciphers pick the side. Over 60% the favourite is the pick, unless the numbers lean three or more against it or the book is wider than six cents, in which case the rule passes and says why. Every pick is locked at the last read before the start and graded when the game settles. Signals, not financial advice.

reading the games…

Toss-ups · the numbers pick, ranked by how far they lean

Clear favourites · the favourite is the pick, unless · next to start first

The full board: every pick, the price tab, the record as it grades

The numbers, live

What the robot has read, and what settled.

Computed in the database from every read the logger has made, refreshed every fifteen minutes. A blank tile means the record has not reached it yet; nothing here is typed in by hand.

Reads logged

—

since the logger's first pass, all time

Markets read

—

every bracket of every ladder

Brackets ruled out

—

by the station's own observations

Markets settled

—

scored against Kalshi's result

The favorite rule

—

per contract, day before, at the ask plus the fee

Same-day favorites

—

per contract, at the ask plus the fee

Queue at 1c

—

median Yes asks resting on a dead bracket

Observation's misses

—

dead brackets that settled Yes anyway

The desk that started this · closed 2026-09-17

Seven weeks of forecasts against the price. The price won on every cut.

This site began as a forecast desk that tried to beat Kalshi's weather prices with public NWS data. It was scored head to head on every settled market, and lost, in public. The full note and its tables are on the receipts page; the measuring is what survived.

11,427

settled markets

211,539

logged reads

−4.02c

per contract, taker

42.4%

of signals paid

The receipts: settled weeks, graded calls, the retired desk's record

The favorite rule buys Yes the first time a bracket's midpoint sits between 65c and 90c, at the ask that was resting plus Kalshi's taker fee, and holds to settlement; per contract is the mean over settled entries, in cents. A week with under 20 settled entries cannot decide anything, and no figure here is a recommendation.

The opening

Everyone repeats who wins in these markets. Almost nobody reads the tape.

The problem

The stories that circulate about prediction markets, the quiet whale, the boring strategy that prints, the edge that anyone could farm, are told from leaderboards and screenshots. The fills, the order books and the trade tapes that would confirm or kill them are public, and they go unread.

What the desk does

A robot with no opinion reads every ladder every hour and writes down what was there. A finder pulls every weather wallet's fills each week, resolves every market, scores and profiles them, and measures what copying each one would have been worth. We read the exchange's own tape from the moment a contract is decided to the close. The price is scored week by week, and our calls are posted with a date and graded when it arrives, misses left up.

How it is done

Built like a desk, not a tip line.

Sources on every claim

fills · tape · order book · observation · settlement · time

Every figure names the endpoint it came from and the moment it was read. The code that computed it is in the open, with tests, and the database views that build this page are in the repository.

We never tell you what to bet

descriptive, not prescriptive

You get the measurement and the receipts. Most of what we find is that an edge belongs to someone else, a market maker's queue or a bot's observation feed, and we say so.

We grade ourselves first

dated call · rule · probability · result

The desk this site began as lost to the market's price and published the loss. The calls page keeps that habit: a call is never edited after it is posted, and an ambiguous rule is graded against us.