The crypto thesis that everyone repeats - the one about onchain assets being composable, sovereign, and productive - mostly hasn't reached prediction markets yet. Polymarket has built one of the cleanest applications of conditional tokens in the wild, and the outcome shares trading hands every day are real ERC-1155s sitting in real wallets. But the moment you hold one, the composability story breaks.
You can't borrow against it. You can't post it as collateral anywhere else. You can't earn yield on it. You can't loop it, hedge it, or compose it into any other DeFi position. It's an onchain asset that behaves like an offline one. You buy it, you wait, you redeem it. That's the whole stack.
This is strange. Every other meaningful onchain asset got a credit layer within months of becoming useful. ETH did. stETH did. Stablecoins did. LP tokens did. Even illiquid NFTs got NFT-fi protocols (with mixed results, but the attempt was made). Polymarket has been live for years, it routes hundreds of millions in volume, and its outcome shares - the most opinionated, conviction-laden asset class crypto has produced - still don't have a working credit layer.
The reason isn't that nobody thought of it. The reason is that prediction market shares are technically harder to lend against than anything else onchain, and most teams that started looked at the problem honestly and walked away.
Lending against ETH is easy in the technical sense. ETH is continuous. It moves up and down. A liquidation engine has time. The probability of ETH going to zero in a single block is essentially zero, and the design of every major lending protocol implicitly assumes this.
Outcome shares don't behave that way. A share priced at $0.72 today resolves at exactly $0.00 or exactly $1.00 in a few weeks. The path there can be sudden. A 30% drop in three minutes on real news is not a rare event in prediction markets - it's a Tuesday. The asset has a deadline. The asset has binary terminal values. The asset can be permanently worthless in a way that ETH can't.
Any credit protocol that wants to lend against outcome shares has to solve four problems at once:
Price discovery on a thin orderbook. Polymarket's CLOB is the only meaningful price reference. Top-of-book is trivially manipulable - you can post one fake bid and move it. So the oracle has to walk the book and weight by depth.
Liquidation that can actually execute. It's not enough to detect that a position is underwater. The liquidator has to be able to sell the collateral on the same Polymarket orderbook that just crashed. Depth that exists in the abstract doesn't help; depth that exists during the liquidation does.
Binary terminal risk. When the market resolves, the losing side is worth zero. If the protocol has been lending against losing shares right up to resolution, the loss becomes bad debt that the lending pool absorbs.
Lender liquidity guarantees. When your collateral can become worthless overnight, your lenders need confidence they can pull capital out before any specific resolution event. If they can't, they won't lend in the first place.
These constraints are tight. They're tight enough that a naive lending protocol applied to outcome shares would lose its lenders' capital within months. This is probably why the big DeFi lending teams - Aave, Morpho, Compound - haven't shipped this. The composition risk is real and the design space is narrow.
I've been looking at PredMart, which is the first protocol I've seen that takes the four constraints above seriously and ships a working answer to each one.
The oracle doesn't read top-of-book. It walks the Polymarket bid book starting from the best bid, takes the next-best, continues until the cumulative size reaches 5% of pool TVL, and signs the volume-weighted average across all those levels. To meaningfully move the oracle, you'd need to post fake bids totaling at least 5% of pool TVL across multiple price levels - real capital at real risk on a public orderbook. If the bid book is too thin to support the full walk, the oracle declines to sign and borrowing on that market pauses. Manipulation isn't impossible, but the round-trip cost exceeds anything you could extract.
The LTV curve isn't flat. It scales with share price - 75% at $1.00, dropping smoothly to 2% near $0.00, with seven price anchors defining the shape. A share at $0.95 (very likely to win) gets close to its full LTV. A share at $0.10 (very unlikely to win) gets almost nothing. This means the protocol automatically tightens risk on the shares that can go to zero, without needing a governance vote per market.
Liquidations run sub-second from a live WebSocket feed of orderbook changes. Health factor drops below 1.0, the engine moves. The same orderbook walk that produces the oracle price produces the size the liquidator can execute, so depth is checked at the moment it matters. A price drop guard - blocking new borrows when a token drops 35% relative and $0.08 absolute within a 3-minute window - kills the most obvious crash-and-liquidate path before it starts.
The interest rate model uses a two-slope kink at 80% utilization. Below the kink, rates are reasonable. Above it, they spike steeply, up to several hundred percent APR. This isn't to rent-seek - it's to make sure the pool always has liquidity for lender withdrawals. When your collateral can become worthless tomorrow, lender exit liquidity is non-negotiable.
Per-market borrow caps are sized off the 25th percentile of 7-day ask-side depth, meaning the protocol only takes on exposure the orderbook can actually absorb during a liquidation. Beyond that, a 5% pool cap per token ensures that even total bad debt on one market caps out at 5% of pool assets.
None of these mechanisms are flashy individually. Together they're what makes the credit layer functional rather than theoretical.
There's another design choice worth flagging for a Mirror audience, because it matters for the people who actually care about this stuff.
PredMart is non-custodial. The contracts are verified on Polygon and deployed at a public address (0xD90D012990F0245cAD29823bDF0B4C9AF207d9ee). Funds sit in the contract, not in a company wallet. No team member can move them unilaterally. Risk parameter changes sit behind a one-way timelock. The protocol was audited by Hashlock, with the full report public.
This matters because there's a competing protocol in the same space - Ultramarkets - that took the opposite design choice. They run as a custodial prime broker, holding user shares directly, with private unverified contracts and self-reported metrics. They've shipped, they have users, and to their credit they've had a CDSecurity audit. But the trust model is fundamentally different: they hold your shares, and you trust them not to mismanage. If that trust ever breaks, the question of who actually owns what becomes a function of their internal records rather than the chain.
Crypto people know how that movie ends. We've watched it enough times. The whole point of putting things onchain is that you don't have to trust the operator. A credit layer for prediction markets that relies on operator trust is a credit layer for a normal product, not for crypto.
Prediction markets are an information asset class. The prices encode real-world probabilities, weighted by capital at risk. They're useful precisely because someone with skin in the game put their money behind a view.
But information quality is bounded by capital efficiency. If the most informed traders are capped at 1x exposure, their information is underweighted in the price relative to what it should be. If long-duration positions lock capital that could otherwise be deployed into shorter-duration informed bets, the signal in those shorter markets weakens. If outside USDC holders can't capture the yield from meeting borrowing demand, the lending pool stays small, which keeps leverage scarce, which keeps the 1x trap in place.
A working credit layer on outcome shares doesn't just help individual traders take bigger positions. It improves the price signal that prediction markets produce. It makes prediction markets more useful as forecasting tools, not just as betting venues.
This is the kind of infrastructure unlock that compounds. Once it's in place, prediction markets stop being a niche and start being a venue for serious capital. Once they're a venue for serious capital, the prices get more accurate. Once the prices get more accurate, the markets become useful for hedging real exposures. Once they're useful for hedging, they get integrated into other DeFi protocols. The whole stack gets denser, more composable, more useful.
The thesis of onchain finance has always been that productive assets, sovereignty, and composability are the things. Prediction market outcome shares have been waiting for their turn. The first protocol that actually solves the technical problem of lending against them is the one that lets the rest of the stack build on top.
The credit layer is the unlock. Everything else is downstream.
PredMart is live on Polygon. Smart contracts are deployed at 0xD90D012990F0245cAD29823bDF0B4C9AF207d9ee and audited by Hashlock. Full docs at predmart.com/docs.