Private access · proven on practice money

An AI that shows its work.

OddsFinder researches real-world prediction markets around the clock, publishes every forecast before the outcome exists, and keeps its losses on the record forever — right next to the wins. No screenshots. No cherry-picking. The proof runs below, live.

Access is by invitation only. There is no purchase button, no request form, and no contact link — if someone showed you this page, they are your way in.

Built like an instrument, not a tip sheet

It forecasts

Independent evidence for every market it touches — multi-model ensembles, live observations, and calibration against years of measured error. When its math and the market's price disagree, that gap is the edge.

It disciplines itself

Hard caps on every position and on total exposure. Fees modeled exactly. Capital goes to the fastest, strongest edges first — and when nothing clears the bar, it sits in cash and says so.

It proves it

Every decision is sealed with a SHA-256 fingerprint before the market resolves. Change one digit of history and the fingerprint breaks. The record can be exported and verified by anyone.

It mostly says no. That's the product.

Every ten minutes, the engine scans thousands of open markets and talks itself out of nearly all of them. A trade happens only when independent evidence, honest costs, and a plausibility check against the market's own price all agree. Yesterday's cascade, live:

0markets scanned
0researched with independent evidence
0cleared the hard filters
0worth betting on

Two years of receipts, graded in public.

Before trusting the model with a single practice dollar, we replayed it against two years of history — what its forecasting models actually predicted on each past day, graded against what actually happened. No look-ahead, no curve fitting. The results live in the database, queryable by anyone with access.

41,499historical forecasts graded against real outcomes
±1.3°Faverage temperature miss across 19 cities
2.5×more accurate than guessing on rain, by Brier score

One real trade, start to finish

August 18: the market priced rain in Boston at 52¢ — a coin flip. Four independent weather models put the real chance far lower. The engine took NO, sealed the decision with a hash, and sized it inside the caps. Boston stayed dry. +$96 settled to the practice book the next morning — and the New York version of the same bet lost $62.92 the same night, which is on the record too. That is what honest edge looks like: not every bet, just the math over time.

One engine. Every market that can be measured.

Prediction markets now price thousands of real-world questions — weather, markets, economics, elections, sport. Anywhere an outcome is objective and the evidence is measurable, the same discipline applies: model it independently, bet only proven edge, record everything. Weather is the proving ground, not the product.

WeatherLive — trading & settling daily
Stock indicesScanning every cycle
CryptoModel built — validating
Economic dataIn design
Sports & beyondOn the roadmap

The arc is deliberate: prove it on practice money in public → open sandboxes where invitees run the engine themselves → live capital, only when the record and the rules say so.

The guardrails are not settings.

Every account runs the same risk rules, fixed by the house. There is no leverage slider, no “expert mode,” no way to get cocky with money — practice or otherwise. You cannot turn these off. Neither can we.

5%maximum of bankroll at risk across all positions, ever
1.5%maximum in any single position — the budget must spread across at least four separate bets
¼ Kellyfractional sizing, scaled down further by measured forecast skill
6%daily drawdown halt — a bad day stops the engine, automatically
0overrides available to any user, at any tier, for any reason

Losses stay on the page forever.

Every forecasting system loses sometimes. Most hide it. OddsFinder prints its losses beside its wins, permanently — because a record you can trust is worth more than a record that looks perfect.

Inspect the full record
SETTLED · LOST

KXHIGHCHI-26AUG19-B78.5

Position
YES × 5
Entry
6¢
Result
-$0.32
6a1c50cfe90c1bc55446

The questions skeptics ask first

Is this real money?

No — deliberately. Everything runs on Kalshi's practice exchange with practice dollars while the record proves itself in public. A real-money tier exists in the design but is hard-locked in the software until the evidence and the lawyers both say go.

What is a prediction market?

A regulated exchange (Kalshi, CFTC-regulated) where contracts pay $1 if a real-world event happens — rain in Boston, an index closing above a level. The price is the crowd's probability. When our independent math says the crowd is wrong, that difference is measurable edge.

Why should I believe the track record?

Because it cannot be edited. Every decision is hashed before the outcome exists; the full record — losses included — exports to a file anyone can audit. A fabricated history would break its own fingerprints.

Why invite-only?

A proving season deserves a small, serious cohort. No paid ads, no funnel, no pressure — access is granted personally, and there is nothing to buy.

Can I blow up my account?

No. Risk limits are fixed by the house and identical for everyone — capped exposure, forced diversification, automatic halts. There is no setting that lets anyone bet like a cowboy, including us.

You must be invited.

OddsFinder is in its proving season: practice money, real markets, public record. A small group is invited personally — there is no request form, no waitlist, and no one to email. If someone shared this page with you, the invitation runs through them. Ask them.

Already invited? Sign in