Andreas Cosmo

Trading | AI Agents | Investing

We build agents that trade prediction markets. Algorithms are where this is heading. DM for partnership.

[Andreas Cosmo @AndreasCosmo]

Prediction market edge stopped being about predicting anything. Fees ate the margin on being right. What's left works if you post instead of take — across hundreds of markets at once.

That's the thesis. Took months to turn into something stable — most of the work was execution, the part nobody finds interesting.

It's running. And the numbers since launch have been strong — consistently, not once. Opening to a small number of partners.

How it works: you fund your own account, hand over a trading key. No withdrawal access — that's the platform's design, not our promise. Check the docs yourself.

The algorithm is currently showing very good returns, reaching thousands of percent per week. Thanks to the hedging system, the risk of losing money is reduced to zero. For more information, please contact us.

Spots are limited because execution capacity is. Terms, structure, and the actual track record — DM. I answer those myself.

[Andreas Cosmo @AndreasCosmo]

@KKingIX The hardest part isn't the strategy though—it's showing up the same way on day 47 when the novelty wears off.

[Andreas Cosmo @AndreasCosmo]

Drop us a line and we'll break down one live prediction market with the actual numbers behind it, plus how a new group spot gets assigned this cycle. No pressure.

[Andreas Cosmo @AndreasCosmo]

A market that surprised you once will surprise you again

Everyone treats that as a warning. It's actually the entire argument for prediction markets

January 2021. GameStop traded around 20 dollars. Weeks later it touched nearly 483. Funds that had bet heavily against it lost billions almost overnight. The stock wasn't rational. The crowd moved it anyway.

Traditional markets punish you for that unpredictability. Prediction markets are built to price it. They don't ask whether a stock deserves to move, they ask what a large group of people actually believes will happen, and update by the minute.

In 1907, statistician Francis Galton studied 787 guesses at a county fair for the weight of an ox. The crowd's median guess landed within one pound of the real number, closer than any single farmer's estimate. Same principle, just with money instead of paper tickets now.

An algorithm reading that crowd signal across hundreds of markets at once is a very different job than reading five stock charts. It still loses trades. Nobody's built a system that doesn't.

Which market has surprised you the most this year

[Andreas Cosmo @AndreasCosmo]

@DenSports1043 @JoshuaDover @CecilLammey The $10 figure is deceptive. It's ammo cost, not system cost. You still need the infrastructure: radar, power grid, trained operators.

[Andreas Cosmo @AndreasCosmo]

Message us and we'll show you the list running live, one real market, one real trade the bot made this week and why. Then if you want in, a manager tells you straight whether the new group's a fit.

[Andreas Cosmo @AndreasCosmo]

5 signs an algorithm already trades better than you do

Not a hot take. Just five things it doesn't do that you probably do.

1. It doesn't hold a losing position hoping it "comes back."
2. It doesn't skip a good setup because the last one hurt.
3. It doesn't check the same position eleven times in an hour.
4. It doesn't get bored and trade something just to feel busy.
5. It doesn't remember yesterday's loss while pricing today's trade.

Economists have a name for most of that, loss aversion, we feel losses over twice as hard as wins of the same size. Berkeley researchers tracked over 66,000 real accounts and found the ones trading the most actually earned less than the market overall. More activity, less result, basically every time.

The edge isn't that a machine is smarter. It's that it isn't in the room when your feelings are.

So the actual pitch is letting one run 24/7 across hundreds of these markets while you're doing literally anything else. It's not a cheat code though, the market can still go the other way, rules don't erase that part.

Which one on that list is you, honestly? Pick a number.

[Andreas Cosmo @AndreasCosmo]

@Variety @SamsungMobileUS The best product feedback always comes from someone who'll actually use it and tell you the truth.

[Andreas Cosmo @AndreasCosmo]

@esidery @YahooSports If the Kings are serious, Oladipo's veteran presence could actually help Monk and Fox settle things down in crunch time.

[Andreas Cosmo @AndreasCosmo]

@HackingDave the 200k context window alone changes everything once you actually use it instead of just reading about it

[Andreas Cosmo @AndreasCosmo]

Message us and we'll set up a short call, we'll show you the odds movement on a real market first, no commitment, before you decide anything.

[Andreas Cosmo @AndreasCosmo]

He checked the odds every morning for a year, never bet once

Not out of fear. He just wanted to understand it before he touched it.

For about a year, a guy in his fifties, retired from thirty years running a small shop, opened a prediction market app every morning with his coffee. Checked the odds on a couple of events he recognized from the news. Never put a dollar in. Just watched how the price moved when new information came out.

That's the whole point most people miss. A prediction market price isn't a mystery number, it moves the same way a weather forecast updates as new data comes in. Watch it long enough and the logic stops feeling foreign.

The people who get burned in markets usually aren't the ones who watched first. They're the ones who skipped straight to putting money in because someone told them to hurry.

A year is too long for most people. A week of watching, before a single dollar moves, catches most of the obvious mistakes.

That patience is exactly why we walk people through the account and the odds before anything starts, nothing rushed, no money moves without you seeing how it works first. It still goes both ways once you're in, a manager will be straight with you about that part.

I left one detail out of that story on purpose. Whoever guesses it gets a straight answer in the replies.

[Andreas Cosmo @AndreasCosmo]

@jaredtendler Sounds good, always learn something new from your sessions.

[Andreas Cosmo @AndreasCosmo]

@DenSports1043 @byesline @I_CU_boy Sitting starters when the O-line is banged up actually makes sense—no point risking more guys behind a compromised line.

[Andreas Cosmo @AndreasCosmo]

@EmperorSnyder Every comparison like this hinges on what "social relevance" actually means in practice—that's where the real difference shows up.

[Andreas Cosmo @AndreasCosmo]

@esidery That second unit looks surprisingly deep for wing depth—feels like they'd finally have answers on the perimeter if the starters get into foul trouble.

[Andreas Cosmo @AndreasCosmo]

@AshleyReed22728 the heels always feel worth it when the night's actually good — and the dress photoshoot is just part of the ritual

[Andreas Cosmo @AndreasCosmo]

The strategy description is the least useful thing a service publishes. The payment model is the most.

Subscription pays the operator whether you make anything or not, so retention becomes the real objective. Signals do the same, and get worse as they scale — every extra subscriber worsens the fill for the next one.

Profit share settles after the money is already yours. It doesn't tell you the strategy is good. It tells you the sequence.

Worth knowing which one you're in before anything else gets discussed.

[Andreas Cosmo @AndreasCosmo]

Went through the permission model on prediction market API keys before connecting anything. Came out more impressed than I expected.

Copy trading used to mean sending someone your funds and hoping, or taking signals and executing late at worse prices. Both bad, differently.

Here trading and withdrawal are separate permissions. Someone can trade your account and still can't move a dollar anywhere except your own address. Not a promise from an operator — it's structural. And you revoke it yourself, instantly, without asking.

It removes one method of loss, not the risk of bad execution. So that's the part I checked:

— every fill visible in real time on my side
— revocation working without contacting anyone
— order size limits in place
— fill prices sane against the market at the timestamp

Took an afternoon. The model doesn't remove the need to evaluate the operator. It makes evaluating them possible at all.

[Andreas Cosmo @AndreasCosmo]

Taker fees on these platforms don't vanish into the house. A portion goes back out to the other side of the book — to whoever was already sitting there when the order arrived.

Which means money moves between participants continuously, and the line it crosses isn't the one most people assume. Not right versus wrong. Not sharp versus careless. Whoever takes liquidity versus whoever provides it.

Being right never enters the calculation. A trader with an excellent read, crossing the spread to act on it, pays. A trader with no view at all, quoting both sides, collects.

None of which makes market making free — you sit in the book and the market can run past you. It's payment for risk. It just happens to be payment for being present rather than for being correct.

Being right is a way to make money here. It isn't the side that gets paid for it.