EDGEDESK is the professional way to trade prediction markets: a real terminal with every order type, one-click Kelly sizing on any market (the stake comes from the model’s probability, not a hunch), and a proprietary signal engine that prices only what it can defend — and shows its work on every call.
| Market | Action | Price | Model | Edge | Book | Closes | ||
|---|---|---|---|---|---|---|---|---|
| This board is drawn live from the engines. None of it is a screenshot or a sample. The rows arrive from the scanner running against the index — and if this page cannot reach it, the board stays empty rather than filling with markets that do not exist. Open the live board → | ||||||||
A model price measured against the halfway point between bid and ask — a price nobody will actually give you — is easy to publish and impossible to trade. Every number this engine puts on the board is quoted at the price the order would actually go in at, after the fee that entry pays, against the size actually waiting there — and it is withheld altogether when the model cannot defend it.
Kalshi built a great marketplace. We built the cockpit that professional traders bolt onto it.
The whole order book — every buy and sell order waiting on the market — both sides at once, the spread (the gap between them) pinned in the middle, size beside every price. Click any level and your order prefills with the size actually waiting there — capped to your risk limits. Every row on the board carries BOOK: the contracts actually sitting at the price you would hit.
Limit, GTC (stays working until you cancel it), post-only (guaranteed to wait in line, never to chase), IOC (fills instantly or cancels itself) — all of it. Plus the Adverse-Fill Shield: set the worst fill you will accept, and if the market trades past it the desk moves your order to a new price or sends a cancel. A cancel is a request — the exchange can fill first.
Take a price someone else posted and Kalshi charges you the taker fee on every fill; post your own price and wait — a maker order — and you pay nothing on most series, but not on all. Every ladder level shows your price after fees both ways, from the fee schedule Kalshi publishes for that exact series, so you stop donating cents on every fill.
Open the demo terminal →Set your take-profit and stop-loss up front, and get alerts that stage an order before you need it.
One keystroke (Cmd-K) pulls up any market, from anywhere on the desk.
Every price level for an event on one screen, so the mispriced ones stand out.
Every trade is logged for you, with the result filled in when the market settles.
See exactly when each position settles and your money frees up.
See how much you have riding on each theme and each event, against the limits you set.
Prices, profit-and-loss and available size refresh on every row, in real time.
Every position is re-checked against the live price on each update — every 6 seconds while the desk is open — and flagged once it has moved 8c.
Your keys, your Kalshi account. We never hold your funds — an order fires only when you click.
The desk isn't only for our signals. Search any market on Kalshi, bring your own thesis, and trade it with the machinery of a professional operation behind you.
Every 2.5 minutes the engine re-runs its numbers on every market it follows against the live quotes, kills the junk with fixed rules, and publishes what survives — an idea that fails a single gate never reaches the board. It hunts EV — expected value: the gap between what a contract is really worth and what the market is charging for it. The probabilities come from models we fitted ourselves, on real history — what is underneath →
The reasoning printed here is the engine’s own, on a market it is offering right now — not an example written for a brochure. When there is no live engine to read from, this card stays empty rather than inventing one.
Most “AI trading” products are a general model asked a general question. Ours was built the other way round: the most capable public frontier models, wrapped in our own proprietary layer — fifteen in-house engines, fitted over months to what actually happened on this one exchange, its patterns and its math.
The models underneath are ours — proprietary, built in-house. They are fitted on real history — 37,027 baseball games with their inning-by-inning line scores reconciled and the umpire who stood behind the plate, 194,380 individual pitching and batting starts underneath them, 7,276 completed football games with the rest days, roof and wind they were played in, and decades of daily weather at the exact stations these markets settle against. The rain record reaches back to 1940; the temperature record is shallower. The table beside this counts them separately, because they are not the same archive. Not a scrape of opinions. The record of what happened.
That took a long time to build, and it keeps building. Every model here earns its place — tested against what actually happened, again and again. And we hold the data to the same standard as the models. When one temperature feed turned out not to be the data these markets actually settle on, we didn’t patch around it — we taught the model to refuse those days entirely. If the data isn’t settlement grade — the same numbers the market pays out on — it doesn’t get to make a call.
Games, 2010–2025, line scores reconciled · Retrosheet
Per-start detail behind every pitching market, 2022–2025 · Retrosheet
Per-start detail behind every batter prop, 2022–2025 · Retrosheet
Completed games 1999–2025, each with its rest days and result, and its roof and wind where the record carries them. The 272 already scheduled for 2026 carry everything but the score · nflverse
Daily precipitation at the stations these markets settle against — 18 stations, the deepest carrying 31,439 daily observations back to 1 January 1940. Coverage is not uniform: nine stations cover ninety per cent or more of the days since, and four stopped recording decades ago · NOAA GHCN-Daily
Daily highs and lows at 22 stations, each series with its own start date. The rain record is the deep one: the longest high-temperature series here begins in 1970, the longest low in 1990 · NOAA GHCN-Daily
Settled values pulled out of Kalshi’s short public window and kept, because the window does not keep them · Kalshi public API
Ours says no far more often than it says yes, and every refusal comes with its reason. An engine that only ever finds opportunities is not an engine — it is a salesman.
Kalshi’s public data keeps settled outcomes for a rolling window of roughly seventy days, and nothing older. There is no multi-year archive to purchase, scrape or reconstruct — not for us, not for a competitor, not for a funded team starting next year.
Depth like that cannot be bought; it is captured, one day at a time, into an archive that is ours alone. The capture runs every day the service runs, and a day it misses can never be recovered. The advantage is not that our archive is already deep — it is that the only way to earn depth is to have started.
The engine doesn't wander the whole board hoping. Its proprietary lane models hunt where mispricing has a cause — a repeatable, checkable reason the crowd's price drifts from the evidence — and it can explain that reason on every single card.
An edge measured in cents survives only when execution costs less than the edge.
Each product stands on its own. They were built as one system: every call the scanner publishes is priced exactly the way the desk executes it — same fee math, same size caps (never more than the market can actually fill), same rule for when a call goes stale.
A small window that lives in the corner of your desk and already knows what you're looking at — the market, the live order book, your position, your risk settings. Ask it anything, or press Run Diligence on any market on Kalshi and get the full underwrite: probability, the edge after fees, sources, and the resolution clause that decides it.
Ask it about any market on the board, in plain words.
It answers from the exchange’s own settlement clause and from sources it can cite. When it cannot answer, it says so — no invented figures, no guessed probabilities. The full diligence run goes further: sources, a probability, the edge after fees, and the resolution analysis behind it.
Off, the way it ships. It starts running, and spending, only when you switch it on.
Prediction markets are thin — each one can absorb only so much money before a signal stops being tradeable. An uncapped signal service destroys its own product with every seat it sells. So we cap ours — and we'd rather turn subscribers away than dilute what they're paying for.
Checkout is not open yet. Leave your email and we send one note, the day seats go on sale. Your first two days are free; the dashboard is live the moment you log in, with the full signal feed and paper-trading mode. Get early access →
Optional. The key is stored encrypted and can only place trades — we never hold your funds; they stay on Kalshi. Prefer to just read signals? Skip this step entirely.
Set your bankroll and Kelly fraction — how aggressively each trade is sized — and every order size adapts. You approve every trade — the engine researches, the desk executes, you decide.
Buy the desk, buy the scanner, or bundle both. It's priced like a professional tool: if the fee optimizer saves you a few taker fees a week — the fee Kalshi charges for filling instantly at the going price — the desk has paid for itself. Seats open when checkout does.
Every plan with the Analyst in it comes with a monthly allowance of credits. A question costs 1, a full diligence run costs 5, and a Watch refresh the same 5 — every control prints its price on its face before you press it. Top-ups are flat-priced packs, and a spend cap is on by default — your subscription is the most you can be charged unless you raise it yourself.
How credits work →Expected value — the difference between what a contract is really worth and what it costs. If a contract has a true 29% chance of paying $1 but trades at 10c, buying it is +19c of EV per contract: over many such trades, that gap is where profit lives, even though any single trade can lose. Our engine estimates that true probability from models we fitted ourselves, subtracts fees, caps it to the size actually waiting to trade, and shows the result. The estimate can be wrong — that's why every one is shown with its evidence, and expires the moment the market moves against it.
One unit of Analyst work, at a flat price per action: a question costs 1 credit, a full diligence run costs 5 — the same price however long the answer runs. A monthly allowance comes with the Trade Desk, top-ups are flat-priced packs, and the spend cap is on from the start — reach it and the Analyst stops and asks before spending another credit. Your research is private either way: it never enters the shared signal feed and never uses one of the 500 Scanner seats. See the whole thing on one panel →
Switch on Watch for a market you have run diligence on and the desk keeps that verdict fresh: it re-runs the diligence daily, and immediately if the price moves 8c against it, so a call you made on Tuesday is not quietly wrong by Friday. Each refresh costs 5 credits — the same as running it yourself — the price is printed on the toggle, and it halts at your spend cap. Watch is off until you switch it on, one market at a time — the desk only spends credits you asked it to.
No — and be suspicious of anyone who says otherwise. We publish theories of mispricing with the evidence behind them, and every call carries a freshness guard that blocks it the moment it goes stale. Markets adapt, edges decay, and any individual trade can lose. EDGEDESK is a research and execution tool; every trading decision is yours.
Never. EDGEDESK is non-custodial: your funds sit in your own Kalshi account, your API key is encrypted and can only be used to trade, and orders fire only when you click. A per-account kill switch disables execution instantly.
That's exactly why the EV Scanner is hard-capped at 500 seats. The cap comes from measuring how much the markets can actually absorb across our live signal set — not picked for marketing — and it can only rise if Kalshi's liquidity measurably grows, never because the waitlist looks profitable. EV is always measured against the size actually waiting to trade right now, and the engine scans thousands of markets to publish the few that clear the gate — so the board is never one crowded name. Any seat count we display is real; when it's full, it's full.
Kalshi is a CFTC-regulated exchange and EDGEDESK is analysis and trading software — not a broker, not a fund, not individualized investment advice. You trade your own account at your own discretion. See the full disclosures below, which every subscriber agrees to before billing begins.
It reads the exchange's own settlement clause, word for word. Every market on the desk shows its full title and Kalshi's published rules text, and the Analyst works from that wording rather than the headline — a title is shorthand, the clause is what actually pays. That is why every verdict quotes the clause it turned on, names the source that settles it, and flags the wording that could go either way.
Frontier AI (the Claude model family) wrapped in our proprietary underwriting system — the part we actually built and the part that matters: our own models, fitted on real history, plus a registry of which source settles each market family, a dynamic-field doctrine — rules for facts that keep moving — that catches the traps naive models fall into, fee-and-liquidity math, and the discipline to kill a weak call. We won't pretend we trained our own foundation model; nobody selling you that is telling the truth either. The edge is the system around the model. Today the system runs on its own numbers — the frontier-model layer that writes the analysis out in prose is the part marked AT LAUNCH, and it is marked that way everywhere it appears.
The desk says so, on the row it affects. The index re-checks every open market on a cycle the service measures and publishes — an hour's sleep plus however long the last pass actually took — and the countdown on this page is that number, not a round one. Live quotes for anything on your board refresh every six seconds straight from Kalshi. If a price is older than that, it is labeled with its real age instead of dressed up as live.
On every row. The board carries a BOOK column — the shares actually waiting at the price you would hit, read straight from Kalshi’s order book rather than averaged or estimated. That is why every edge we publish is capped to the size actually waiting to trade: a +9c edge on 109 resting shares and the same +9c on 820 are not the same trade, and the board tells you which one you are looking at before you open the ticket.
No — each stands alone. But they were built as two halves of one workflow: every scanner signal is priced with fees already subtracted and capped to the size actually waiting, and the desk is the instrument that executes exactly that way — one click, Kelly-sized (stake scaled to the model’s probability), fee-routed, freshness-guarded. If you trade the signals anywhere else, expect to hand a few cents of each edge back to fees and slippage — the price moving against you as your order fills. That's the honest case for the bundle.
Checkout opens with the founding tranche — until then the queue is the only way in. Once you are in, your card is billed at the end of the 2-day trial unless you cancel — one click, no emails, no retention games. Cancel mid-cycle and you keep access through the period you paid for. Get early access →
Three ways: the terminal (a genuinely faster way to execute on Kalshi), reasoning you can read (every call carries its evidence chain and a read of the exact clause that settles it, not just a number), and guardrails (freshness guard, fee-aware routing, sizing capped to what’s actually waiting — the desk protects the trade, not just picks it). Screens give you prices; picks give you trust-me. We give you a desk.
One hundred founding seats. When they are gone, the next tranche opens higher. That is not fake urgency; it is how capacity works.
EDGEDESK identifies theories of potential mispricing in prediction-market contracts. These estimates may be wrong, may reflect no true edge at all, and past performance does not guarantee future results. Trading event contracts involves substantial risk of loss and is not suitable for everyone. Never trade with funds you cannot afford to lose.
EDGEDESK provides research, analytics, and execution software for your own self-directed trading. Nothing on this platform is individualized investment, financial, legal, or tax advice, and no advisor-client relationship is created. All model outputs, probabilities, "EV" figures, and verdicts are automated analytical estimates, provided for informational purposes only.
Every order is initiated by you, in your own exchange account, under your own exchange agreement. EDGEDESK is non-custodial, never holds funds, and takes no responsibility for trading outcomes, exchange availability, data errors, or execution results — except for fraud, willful misconduct, and the rights you hold under consumer-protection law, which section 22 of the Subscriber Agreement does not limit and neither does this. By subscribing you agree to the full Subscriber Agreement and Risk Disclosure, including binding acknowledgment of the above.
This page says EDGEDESK is built by people who trade these markets, and it says the seat cap exists because the size waiting to trade in these markets is finite. Both are true, and together they mean everyone acting on the same information — including us, to whatever extent we trade it — is competing for the same fills. The rule we hold ourselves to: nothing is traded on a model output before that output has reached every seat entitled to see it. EDGEDESK charges its subscribers for the Service — subscription fees and flat-priced Analyst credit packs. Whether anything else reaches this operation in connection with your trading is a question of fact about the arrangements actually in force — the answer lives in the disclosure itself, not in a claim made here. The conflicts disclosure in full →
Any performance-related figure shown anywhere on this platform is labeled, unedited, and never cherry-picked; hypothetical figures (such as "EV identified") are identified as such.