how it works

01overview

Every overcollateralized lending market — Aave, Morpho, Spark, and the margin engines behind perpetual DEXes like Hyperliquid — has the same problem: when a borrower's collateral falls too close to their debt, someone has to step in, repay that debt, and take the collateral off the protocol's books before it becomes a bad debt the protocol eats. Protocols can't do this themselves, so they pay an outside party to do it. That payment is a fixed discount on the seized collateral.

REAP is the outside party, pooled. Instead of a single searcher running private infrastructure to win those liquidations, depositors pool USDC into a vault that bids them collectively and splits the proceeds pro-rata. You are not lending and earning interest. You are holding a claim on a stream of liquidation discounts.

02the discount

A borrower's position is healthy while the value of their collateral, marked down by a protocol-set risk factor, still covers their debt. That ratio is the health factor:

When price moves push HF to 1 or below, the position is open for liquidation. A liquidator repays some of the debt and, in exchange, receives collateral worth more than what they repaid — the extra is the liquidation bonus (also called the incentive or penalty). The bonus is the protocol's price for getting the bad debt cleared quickly; it is set in advance and is the same whether the liquidator is a bot or a vault.

03the capture loop

When the vault wins a liquidation it does four things in one transaction: repay a portion of the debt, receive the collateral at the bonus, and then unwind that collateral back to USDC. The gross profit is just the bonus applied to the debt the vault clears:

That number is gross, not realized. Two costs sit between the seized collateral and a USDC balance the vault can show: the price impact of selling the collateral (large seizures move the market against you), and the gas to land the transaction. Net of those:

The unwind term is why size and liquidity matter more than headline bonus: a 10% bonus on an illiquid asset can net less than a 5% bonus on something with deep markets. REAP routes the unwind across venues to keep s low, which is most of the edge over a naive liquidator.

04where returns come from

Across all the venues the vault monitors, some notional amount of collateral gets liquidated over any period. The vault doesn't win all of it — it competes with independent searchers and other vaults — so it captures a fraction at the average bonus, after costs:

Divide revenue by the capital backing it and you get the depositor's return for the period:

This is the whole thesis in one line. Return scales with liquidation volume V and the vault's win rate w — not with a borrow-lend spread, not with an emissions schedule, not with anything that drips evenly. There is no yield when nothing is being liquidated.

05why returns are lumpy

The term that drives everything, V, is not steady. A position only liquidates when a price move is large enough to push its health factor through 1. Rearranging the condition from section 02, a long position is liquidated once the collateral price falls by roughly

On a quiet day, almost nothing crosses its Δ*, so V ≈ 0 and revenue is near zero. In a deleveraging cascade, a sharp move liquidates a band of positions at once, those forced sales push prices further, which trips the next band — and V spikes by orders of magnitude in hours. Revenue tracks realized volatility, and volatility arrives in bursts.

06why the lock-up exists

To win a liquidation you need USDC available at the instant it happens. The vault can only deploy what's sitting in the book, so the amount it can capture in any window is bounded by its deployable capital:

The problem is timing. Volatility is exactly when L_avail is largest and when nervous depositors most want to pull funds. If withdrawals were instant, the book would drain at the precise moment the opportunity appears, collapsing K and forfeiting the capture. A 4-day withdrawal queue decouples the two: it holds K stable through a cascade so the vault can keep bidding while everyone else is panicking. The lock-up isn't a convenience tax — it's what makes the strategy investable at size.

07the buyback

A fixed slice of every realized capture is routed to buy the REAP token on the open market, \( \text{buyback} = \varphi \, R \), where φ is the protocol's buyback share of net revenue. The buyback is funded only by actual realized discount, so it has the same lumpy, volatility-linked shape as everything else — it is a function of R, not of token emissions or a fixed schedule.

08parameters

The symbols above split into two groups: protocol constants the vault sets, and market quantities it can only measure. Live measured values are shown as n/a here and reported on the vault dashboard when data is available.

09assumptions & risks

The model is only as good as its terms hold up. The honest failure modes:

• Win rate is not guaranteed. Liquidations are competitive. Well-capitalized searchers with private orderflow can outbid the vault, driving w — and therefore returns — lower than the mechanism allows in principle.
• Slippage scales with size. As TVL grows, each unwind is larger relative to available liquidity, raising s and lowering η. The strategy has a capacity ceiling; more capital does not mean proportionally more return.
• Long flat stretches. Calm markets can mean near-zero revenue for weeks or months. The payoff is real but irregular; there is no contractual yield.
• Smart-contract and oracle risk. The vault interacts with external lending markets and price feeds. A bug, an oracle failure, or a venue exploit can cause loss independent of the strategy working.
• Bad debt. In a violent enough move, seized collateral can be worth less than the debt repaid by the time it unwinds — a liquidation can lose money, not make it.

None of this is a reason the mechanism doesn't work; it's the set of things that determine how well it works. Read it before sizing a position.