questions ✓

If it works, why sell seats instead of just betting it?

Honestly: this took a long time to build, and it costs real money to keep running — servers, market data, and the harvesting and compute behind every model on the board. Seat revenue covers that. It means the machine pays its own upkeep instead of eating into the bankroll that trades it, and it means the next model gets built on a budget rather than out of last month's winnings.

It also costs the same to run whether one person reads its output or five do, and the fifth reader does not move a price. That is the part people have backwards: a seat is not taken instead of a position. It is the only part of this that does not compete with the positions for room.

The capacity limit is real, and it is why there are 5 seats rather than unlimited. That number sits below the level where combined member orders could move the prices we are taking, and enrolment freezes on its own if fill quality slips. Unlimited access to a small-market edge is either a fake edge or a destroyed one.

What does a seat actually get me?

Alerts the moment a pick fires — market, side, maximum price, size. Past the max price, skip: a worse price is a different bet. Every pick is graded against the venue's own resolution and published, winners and losers alike.

Is this real money or paper?

The entire published record is paper-graded: forward-tested at real prices at alert time, but not realised P&L — and labeled paper on every surface. Anyone who blurs that line is telling you something about themselves.

Why would these markets be beatable at all?

Because we pick the battlefield. Markets priced by professionals track the sharpest books in the world within a cent or two — we measured that ourselves, venue by venue, and we don't trade them. We target markets where recreational money sets the price: real volume, real fills, crowd attention deciding the number.

The crowd overpays for favorites, for excitement, for the famous side — and those habits are measurable, repeatable pricing errors. Put plainly: we are not competing with anyone doing this as a business, as best we can avoid it. We're picking on the people playing for fun.

What is the process for building a model?

A lab, in stages. Most candidates die in one of them.

Harvest. Collect the raw history first, keep it untouched. A missing layer doesn't weaken the answer — it silently changes the question.

Clean and screen. Reconcile the data, then run wide screens before any theory — a thesis invented first finds its own evidence.

Analyse. The conditions that would kill the hypothesis are written before the test runs. Results are judged on independent events — correlated picks on one game are one bet in several hats.

Out-of-sample. Hold-outs, era splits, price sensitivity, artifact checks. This stage exists to kill false edges, and it kills most candidates.

Paper. Survivors run live-priced with no money for a pre-set number of plays — and must agree with their own backtest. Divergence is a failed test, not a slow start.

Live. Only then. Everything that died stays published.

How do you keep the edge from decaying?

We don't prevent decay — we measure it. Every lane carries pre-registered metrics it must keep clearing; regression means demotion back to paper, still graded but no longer alerted. Demotion is the normal outcome, not a crisis.

What we manage is the blended return across live lanes, not any one model — lanes fade at different times, and defending a single dead lane is how services die. Fill quality is monitored too; enrolment can freeze on its own.

Where do I place these bets?

Prediction-market venues — for US members mostly Kalshi, and largely Polymarket US. Whether a venue is available where you live is yours to check; we don't advise on it.

Can I see the record before I buy?

The front-page roster shows every live market class, its record, and the last fourteen days, updating on its own. Alerts also post a public fingerprint the moment they fire, so pick timing can be verified independently.

Who is Kelly? Is this actually run by AI?

Yes — and it's disclosed everywhere because it's true. The system runs grading, publication and delivery; humans check the work and make the calls that matter. Kelly is the operator's voice, not a person, and is never presented as one.

What does it cost — and how much bankroll makes it worth buying?

Founding seats are $150/month, or $350 quarterly (save $100, listed first). The published ladder rises for later tranches; a seat keeps its price for as long as it stays active.

Here is the math nobody else in this business will show you. The fee is flat, so whether a seat pays for itself depends on your unit size. At the feed's historical rate, a bankroll of at least $2,500 across venues (betting ~1% a pick) clears the fee comfortably in an average month. Below roughly $1,500, the fee is likely to eat your edge — we would rather you not buy a seat than buy one that can't pay for itself. The application asks your size for exactly this reason, and we decline applicants the math doesn't serve.

Refunds, cancelling, and getting removed

All sales are final. We do not offer refunds — for any reason. You can cancel at any time and keep access until the end of your paid term. Seats can also be ended from our side at our discretion — the terms are explicit about that. Read the membership agreement and the risk disclosure before you buy. They are short on purpose.