$836M in losing bets exposed Polymarket's structural flaw. Here's what happens when prediction markets actually pay.
In January 2026, over $836 million was wagered on a single Polymarket event: the 2028 Democratic Presidential Nominee. 44 candidates. Hundreds of thousands of bets. The biggest prediction market in history.
A market that size raises a question the book would rather nobody asked out loud. When 43 of 44 outcomes lose, where does all that losing money go?
On Polymarket, it does not go to the people who got it right. It cannot. The answer is baked into the share model.
The Math Behind the Gap
The difference comes from a fundamental design choice in how winning bets get paid.
Polymarket's model: Fixed $1 ceiling.
Polymarket uses a binary share system. You buy shares at less than $1. If you win, each share is worth exactly $1. The maximum payout is always $1 / entry_price. It doesn't matter if $800 million was bet on the losing side — your upside is mathematically capped.
This is simple. It's clean. And it leaves enormous value on the table.
Basis's model: Winner takes the losing pool.
On Basis, prediction markets work differently. All money bet on losing outcomes flows into a shared pool. When the market resolves, winners split that entire pool proportional to their shares.
The formula:
Your Payout = Your Shares / Total Winning Shares × Total Losing Pool
No ceiling. No cap. If the losing pool is large and few people bet on the winning outcome, the payout runs well past the point where a fixed $1 share would have stopped.
Real Data, Real Markets
We didn't test this on toy examples. We replayed 111 settled Polymarket markets, every trade on every outcome and $422 million of notional, through the Basis engine, paying out every position under our rules instead of the book's.
The result: 61% of winning wallets, holding 71% of the winning money, would have finished ahead of what the book actually paid them. Across the 77 markets where a longshot won, the winning side would have taken 1.33x more out of the pool. Early money did best of all: money arriving in the first 5% of a market's life would have been paid 1.85x the book on the typical field, and 4.95x on fields of 31 or more outcomes, where the pool beat the book in every single market.
Two boundaries matter, because they are where the advantage actually lives:
- The pool beats the book up to 70.5c. Above that price you are paying for a near-certainty, and the fixed ceiling is the better instrument. Below it, the pot pays more.
- The window is 71 to 78 percent of the pot. That is the band where the pool takes the largest share, and where the gap against the book is widest.
Parlays make the same point from another angle. A Basis parlay — built, and deploying at Phase 2 — returns 98.5c on the dollar. The same parlay booked in New Jersey returns 75.8c.
The pattern underneath all of it: lower probability outcomes produce larger absolute payouts on Basis, because the losing pool grows while the winning pool shrinks.
Why This Matters for Agents and Traders
Here's where it gets interesting. A human trader might look at this and think "nice, bigger payouts" — and they're right. But an AI agent looks at this and thinks completely differently again.
Agents can model payout surfaces in real time.
An agent with access to the Basis SDK can call getPotentialPayout() on any market, for any outcome, at any moment. It can map the entire payout surface — every combination of bet size, outcome, and timing — and find the optimal entry point.
On Polymarket, there's nothing to optimize. The payout is 1/price. That's it. On Basis, the payout is a function of pool dynamics, and pool dynamics change with every trade. That's a rich optimization surface — exactly the kind of problem agents excel at.
Agents can create markets, not just bet on them.
On Polymarket, agents are bettors. That's the only role available. Buy shares, hope you're right, collect if you win.
On Basis, agents can also be market creators. Deploy a prediction market on any event. Earn 20% of all trading fees on that market — forever. The market creator doesn't need to bet at all to earn.
Consider what this means: an agent can scan the news cycle, identify high-interest events, create prediction markets for them, and start earning fees within minutes. It's not gambling — it's building financial infrastructure on demand.
Humans and agents alike can compound across multiple products.
Here's a strategy that's impossible on Polymarket but native to Basis — available to any operator, human or autonomous:
- Create a prediction market on a trending event (earn creator fees)
- Buy outcome tokens on your highest-conviction bet
- Borrow USDB against those outcome tokens at 100% LTV
- Deploy the borrowed USDB into another market or trade
- Earn creator fees + prediction payout + trading profits + loan arbitrage
- Reinvest everything into the next cycle
That's six revenue streams from a single starting position. The agent isn't just betting — it's running a capital-efficient business across multiple financial primitives.
On Polymarket, you bet. On Basis, you operate — whether you're a human trader, a creator, or an autonomous agent.
The Agent Stack Gap
Let's compare the complete toolkit available to agents on each platform:
| Capability | Polymarket | Basis |
|---|---|---|
| Place bets on outcomes | ✅ | ✅ |
| Create new markets | ❌ | ✅ (earn 20% fees) |
| Launch tokens | ❌ | ✅ (Stable+, Floor+) |
| Take loans against positions | ❌ | ✅ (100% LTV, no liquidation) |
| Access vault yield | ❌ | ✅ (wSTASIS vault) |
| Use leverage | ❌ | ✅ (dynamic, floor-price based) |
| On-chain agent identity | ❌ | ✅ (ERC-8004) |
| Earn creator fees | ❌ | ✅ (20% of trading fees, forever) |
| SDK with full API coverage | Partial | ✅ (595 methods, Python + TypeScript) |
| Composable strategies | 1 (bet) | 18 strategies across 4 decision trees |
| Community & growth tools | ❌ | ✅ (content, posting, community management) |
| Agent-human collaboration | ❌ | ✅ (delegation pattern for blocked steps) |
Polymarket is a prediction market. Basis is a financial operating system.
The difference isn't incremental — it's categorical.
Why Bigger Payouts Aren't the Only Story
The payout comparison makes a great headline, but the deeper insight is about what agents can do with those payouts.
On Polymarket, an agent that wins a bet gets USDB. End of story. There's nothing to reinvest into, no compounding mechanism, no additional products. The agent has to find another platform to deploy its winnings.
On Basis, an agent that wins a prediction can immediately:
- Stake the proceeds in the wSTASIS vault (earn passive yield)
- Use them as collateral for a loan (borrow more USDB)
- Launch a token backed by the position (earn creator fees)
- Create another prediction market (earn more creator fees)
- Fund leveraged positions on other tokens
The earnings loop is closed. Capital never has to leave the ecosystem to find its next productive use. For an agent optimizing for total return across time, this is the difference between a one-shot game and an infinite game.
The Numbers in Context
Polymarket has proven something important: prediction markets work. People — and agents — will bet on outcomes at massive scale. $836 million on a single political event is proof of concept for the entire category.
But Polymarket's architecture was designed for a simpler time. Binary shares capped at $1. One product (betting). No composability. No creator economy. No agent tools beyond basic API access.
The next generation of prediction markets — and DeFi more broadly — will be built for the participants who are growing fastest: AI agents. And those agents don't want a slot machine where the maximum payout is hard-coded.
They want a business.
Try It Yourself
The math is open. The SDK is built. The contracts are live on BNB Chain.
from basis import BasisClient
client = BasisClient.create(private_key="0x...")
# Check any market's potential payout
payout = client.market_reader.get_potential_payout(
router, market_token, outcome_id, shares, estimated_usdc
)Three lines from zero to payout data. That's agent-native finance.
Basis is the agent-native DeFi layer on BNB Chain. A suite of more than 20 smart contracts, deployed in phases. 595 SDK methods across TypeScript and Python. The financial operating system for agents and traders. launchonbasis.com
Basis Team
Published Apr 2, 2026