Live · Kalshi · Polymarket · sports lines

Autonomous trading desks for markets that settle on facts.

Cat3na reads prediction markets, sports lines, crypto, tokenized assets and equities every ten minutes and prices each one itself — then trades only the small share where its number and the market's disagree enough to be worth it.

Your wallet, your limits, and every decision written down before it is acted on.

A note now. One more when desks open.

See what it refused

Of everything it read this cycle
—
was refused before it reached an order — thin edges, stale prices and markets not worth the risk.
The Cat3na command deck: paper mode, day 12, eight trades awaiting review, with the account's live value and amount at risk.

Markets the platform connects

Polymarket
Kalshi
ESPN lines
Base
Polygon
Arbitrum
Equities
ETFs
Tokenized treasuries
Perpetuals
FX
Commodities

One desk. Every market you can settle.

Most platforms ask you to pick a market and then sell you software for it. Cat3na is the other shape: one brain, one ledger, one gate — and as many venues as you point it at.

Adding a market class is an adapter. It is not a second product, a second bill, or a second set of rules about what you are allowed to trade.

01Polymarket · Kalshi

Prediction markets

Every contract that resolves to a fact. Rates, elections, policy, weather, anything with a settlement date and an answer.

02ESPN lines · match markets

Sports & event lines

Moneylines, totals and match markets, read against the game underneath them rather than against the crowd holding them.

03Base · Polygon · Arbitrum

Crypto & on-chain assets

Momentum and rebalancing executed through your own smart wallet. Keys stay yours; the desk never takes custody.

04Cat3na custody rail

Tokenized real-world assets

Tokenized treasuries and yield notes allocated on the custody rail, with the same position sizing as everything else.

05Single names · baskets

Equities & ETFs

Single names and baskets, sized by the same Kelly fraction and stopped by the same twelve checks.

06Continuous venues

FX, commodities, perps

The same brain pointed somewhere else. A new market class is a new adapter, not a new product.

A desk is a signal engine.

Each desk reads its markets on a schedule and proposes trades. Your money never sits inside one — positions are sized into your own account, and the desk only ever holds an opinion.

Reads on a schedule
Every cycle, every connected venue.
Proposes, never places
An order needs the gate and, in review, you.
Scored against settlement
Hit rate and Brier, computed from outcomes.
The desks screen: the Prediction and Sports desk running in paper mode, 1,701 markets analyzed, 262 cleared and 1,439 declined, with its settled record and Brier score.

Five layers. Each one can say no.

A model that only knows how to say yes is not an intelligence system. It is an opinion with a schedule.

Every layer below holds a veto, and every veto is recorded. What survives all five is the only thing that ever becomes an order.

Read this cycle

—

candidate markets

Cleared the edge bar

—

of what was read

Gate state, right now

—

checks passing, live

01of 05The brain

It reads every market and states a number.

For each contract it produces its own probability, independent of the price. Not a signal, not a rating — a number that can later be checked against what actually happened.

The ranked edge signals table: real markets with the market probability, the model probability, the edge between them, confidence, quality score and gate status.

It throws away most of what it just read.

Candidates are ranked by effective edge × quality. Thin edges, stale prices and low-quality markets are dropped here — before anything reaches a risk decision.

A desk showing 1,701 markets analyzed against 262 cleared, with the line "the gate lets through 15.4% of what it reads — discipline, not volume."

Twelve checks. All twelve, or nothing trades.

Exposure, correlation, drawdown, venue health, data freshness, kill-switch state. Live trading unlocks only when every check passes, and the gate is read from the API — including the ones it fails.

The Prediction and Sports desk detail: 1,905 markets analyzed against 291 that cleared the gate — 15.3% — with the Brier score and the minimum edge and Kelly fraction it enforces.

Every decision is written down before it is acted on.

Append-only. Each proposal, refusal, fill and settlement is recorded with what was known at the time — which is what makes a track record something a stranger can audit instead of something you have to trust.

The backtest and verification screen: 185 settled positions sorted by how well each one can be proven.

Your brand, your limits, your kill switch.

Name, mark, accent, venues, Kelly fraction, per-desk caps and the stop that halts everything. Configuration, not a fork — there is no per-tenant code anywhere in the system.

The operator studio: white-label brand, integrations and revenue settings that apply to the whole console instantly.

Not a mockup. The running app.

Every screen on this page is a capture of the console running against the live API — real markets, real prices, real refusals. Including the screens that say a thing is not built yet.

Positions

The whole life of an order, not just the fill.

Entry, current mark, size, P&L and every state change in between. Open a row and you get the fills and the state history that produced it.

Full lifecycle
Proposed → gated → queued → filled → settled.
Marked continuously
Against the venue price, not against your entry.
The positions book: every position with market, desk, side, source, entry, current price, size, P&L and state.

Copy trading

A leaderboard of winners tells you nothing.

Real wallets, scored from every fill we can observe — including the ones losing money. Then re-ranked by a Wilson lower bound, so a trader with nine lucky wins stops outranking one with two hundred honest ones.

Observed, not self-reported
Scored from on-chain fills.
Re-ranked for sample size
Luck falls down the list on its own.
A trader profile in the console: return on risk +146.4% on $1.5K risked, +$2.2K realized rather than on paper, and a 92% win rate.

Performance

Two questions. Honest, and compounding.

A reliability curve answers whether the brain’s probabilities mean what they say. An equity curve answers whether that is worth anything. Most platforms show you the second and hope you never ask for the first.

Calibration, plotted
Stated probability against observed frequency.
Brier, not win rate
A score that punishes confident and wrong.
The performance screen: a reliability chart and an equity curve over 30 days in paper mode.

The rest of the console.

Four more screens, shown as they are — the money already returned, the knobs that bound it, the loop explained before you fund anything, and the five steps that put your brand on all of it.

Committed capital, what is at risk this second, and what has already settled. Realized means realized — money the copies returned, not an open position marked in your favour.

The following screen: $9,500 committed across four sources, $109.37 at risk in open copies, and +$3,017.38 realized from settled copies.

We publish the checks we are still failing.

The twelve-point gate on this page is read from the live API on every load, including the checks that do not pass. Nothing here is a screenshot of a better day.

A platform that only shows you its wins is not showing you a platform. It is showing you a brochure.

Read the record

A track record is only worth what a stranger can check.

Anyone can screenshot a good month. The question is whether the number was computed from settlements you can go and look at, or from a spreadsheet nobody has ever seen.

Below is the live gate, failures included, and the observable universe the copy leaderboard is drawn from.

Sorted by how well it can be proven.

Every settled position, ordered by the strength of its evidence rather than by its size. The trades at the top are the ones you can verify without asking us anything.

The backtest and verification screen: 185 settled positions sorted by how well each one can be proven.

Start on paper. Everybody does.

A sandbox desk is real signals, real prices and real refusals against no real money. It is the same console operators use — nothing is simulated except the balance.

We are not publishing a price band yet, because billing is not built and a price we cannot collect is a number you should not trust.

Sandbox desks are opening in waves rather than all at once. Leave an address and you get the next one.

01

Your brand on the console

02

Every connected market class

03

The twelve-point gate

04

Append-only decision ledger

05

Copy trading and leaderboards

06

Kill switch and per-desk caps

A note now. One more when desks open.

FAQ

The questions worth asking.

Prediction markets and sports lines are reading live today, through Polymarket, Kalshi and event lines. Crypto and on-chain assets, tokenized real-world assets and equities are wired as desk types on the same brain, ledger and gate. Every desk on the platform runs on paper until it clears all twelve checks.
No. Positions are sized into your own account and on-chain execution runs through your own smart wallet. A desk holds an opinion, not custody.
Twelve checks — exposure, correlation, drawdown, venue health, data freshness, kill-switch state and others — that must all pass before live trading unlocks. It is read from the live API on this page, including the checks currently failing.
Because they fail. A gate that only ever renders green is a graphic, not a control. Publishing the real state is the only version of this claim that means anything.
Name, mark, accent, venues, Kelly fraction and per-desk caps are configuration. There is no per-tenant code anywhere in the system, so your console is not a fork that drifts from ours.
From observable fills, not self-reported returns, and it includes wallets that are losing money. Rankings use a Wilson lower bound so a small lucky sample does not outrank a large honest one.
It is auditable, which is a different and more useful claim. Every proposal, refusal, fill and settlement is written to an append-only ledger with what was known at the time, and the backtest screen sorts settled positions by how well each can be independently proven.
Yes. The kill switch halts every desk immediately, and it sits in the console rather than behind a support request.

Watch it work on paper first. Then decide.

A note now. One more when desks open.

Back to the top