Skip to main content

Risks

This page is a structured list of things that can break or behave badly in the protocol, organized by surface. It focuses on concrete properties of the current code and design that users, integrators, and auditors should understand.

Oracle risks​

Stale or failing collateral oracles​

PriceFeeds tolerate oracle latency up to each source's staleness threshold (25 or 28 hours, by source). A stale or failing ETH/USD or BTC/USD primary (Chronicle) makes the feed switch to its Chainlink fallback for good. If the fallback also fails, or any other input of a derivative feed fails (a market oracle, a rate provider, or PAXG's XAU/USD reference), the feed shuts the branch down through shutdownFromOracleFailure() on the next price read (see Collateral oracles).

Effect on user: the branch shuts down. Pricing continues from the last good price (or base price × canonical rate for a derivative whose market leg failed). New borrowing and collateral changes revert and liquidations stop; troves can still be repaid and closed, and anyone can redeem RD against them on the oracle-failure discount schedule, which reaches ~100% after 14 days (see Collateral shutdown).

LST canonical-rate divergence​

For rETH and weETH, the feed compares the market price (ETH/USD × the LST/ETH oracle) with the canonical price (ETH/USD × the token's exchange rate) and uses the lower of the two, except that a redemption read within 2% uses the higher. wstETH is priced from the stETH/USD market price × the wstETH rate; only a redemption read within 1% of ETH/USD uses the higher of stETH/USD and ETH/USD.

Effect on user: during an LST depeg, the branch prices the collateral at the lower market value for everything — ICR, liquidation and redemption. Borrowers on that branch see their ICR drop with the depeg and can be liquidated, and redeemers receive collateral valued at the depegged market price.

RD market pool drain​

MarketOracleAggregator reads the RD/USDC and RD/USDT Uniswap v4 pools. A pool stops contributing a price when its 24-hour liquidity falls below 250,000 RD-equivalent, and the price is not accepted when the contributing pools hold less than 500,000 in total (or RD supply is below 5,000,000). Thinner but accepted markets reduce the controller's liquidity authority, down to a 20% floor.

Effect on protocol: while the price is not accepted, RateParControl skips its control steps, so rate and par stay at their last committed values and the controller can't respond to market moves until liquidity returns. If a pool stays empty or badly imbalanced for 7 days within a 21-day window it shuts down permanently; if both pools shut down, the controller latches its terminal shutdown and rate and par decay toward their biases (see RD Market Oracle).

Same-block manipulation​

Each pool oracle reads only windows that end MIN_OBSERVATION_AGE = 60 seconds in the past, over a 24-hour TWAP. A flash-loaner moving the pool price in the same block as the read cannot poison the controller because the controller does not consume same-block ticks.

This is a known design defense, not a residual risk. Same-block manipulation is mitigated.

Liquidation risks​

SP empty during cascade​

If a price crash triggers a cascade of liquidations and the Stability Pool runs dry, the protocol falls back to redistribution. Every active trove on the branch inherits a pro-rata slice of the liquidated debt and collateral. Surviving troves' effective ICR drops.

Effect: surviving borrowers can be liquidated in subsequent cascades they didn't trigger. This is intrinsic to the redistribution mechanism; same risk profile as Liquity v1.

Negative-yield SP deposit during crash​

If collateral price drops below the SP's effective offset price (e.g. a 20% drop in seconds), depositors absorb troves at a loss, collateral received is worth less than RD burned.

Mitigation: depositors should size their position with the awareness that a black-swan crash can produce a nominal loss.

Below-MCR trove not liquidated​

The protocol can carry an undercollateralized trove for some time if no one calls liquidate(...). Anyone can do it, but if the gas economics don't favor a liquidator (very small trove, congested chain), liquidation may be delayed. The trove's debt is still RD-denominated, the protocol carries the loss until liquidation completes.

Mitigation: keepers, bots, and front ends are incentivized via the 200 RD gas comp + 0.5% collateral kicker. New Stability Pool deposits also first try to liquidate up to 10 troves the pool can fully absorb.

Redemption risks​

Redemption revert​

A single branch reverting during Aggregator.redeemCollateral reverts the entire transaction. This is an operational risk for redeemers and integrators: a bug or denial of service on one branch can stall global redemption. Normal redemption also reverts while any basket branch is eligible for shutdown (TCR below SCR, oracle failing, or collateral underbacked), and when a branch that hasn't run out of redeemable troves fills less than 90% as completely as the best-filled branch (for example, because its partial-redemption hint went stale).

Mitigation: shutdown of the affected branch removes it from the basket; anyone can trigger it with Aggregator.tryAllShutdown() or tryShutdown(tm). Fresh hints from GlobalHintHelper avoid the uneven-fill revert.

Bootstrap-period absence of price floor​

For the first 14 days, redemption is disabled. RD's price floor mechanism doesn't exist during this window. The rate controller is also frozen during bootstrap: rate and par hold at their bias values, so nothing in the protocol responds if RD trades meaningfully below par.

Mitigation: initial liquidity should be sized assuming no redemption floor and no controller response during bootstrap.

Quota exhaustion​

The redemption quota refills with time (4% of system debt per day, at most 1% available at once). A large redemption can deplete the bucket, and a subsequent request larger than what has refilled reverts. There's no admin override.

Mitigation: read Aggregator.redemptionQuotaAvailable() before submitting.

Controller risks​

Stale-poke after long inactivity​

If RateParControl.update(...) has not been called for many days, the accumulated dt is capped at MAX_CONTROL_DT = 12 hours. The controller will not apply a single huge step; it takes subsequent updates to fully catch up.

Effect: in deep tail scenarios (no one updating the controller), rate and par lag the true equilibrium.

Mitigation: keeper rewards on controller updates (KeeperRewards.updateWithReward()) incentivize regular updates. Borrower operations, liquidations and Stability Pool deposits also refresh the controller through the branch drip.

Asymmetric gains​

The rate loop's under-peg proportional gain is 4.5× its over-peg gain (the integral gain is symmetric). This is intentional because RD below par is the more dangerous mode, but it means the controller responds faster to RD below $1 than to RD above $1. During a sudden over-peg event, expect the rate to fall less aggressively than it would rise for the same deviation below peg. See Peg control.

Immutability​

The contracts are immutable and effectively ownerless: no admin keys, no upgrade path, and no owner able to change any parameter after deploy. Nothing below is upgradeable or admin-changeable:

  • Aggregator, RDToken, MOAIToken, RateParControl, MarketOracleAggregator, the RdStableSwapHook pools, and GlobalFeeRouter are fully immutable; RateParControl's controller constants are hardcoded.
  • Per-branch core: BorrowerOperations, TroveManager, Liquidations, Redemptions, Rewards, StabilityPool, InterestEngine, SortedTroves, and all pool contracts.
  • Per-branch parameters (MCR / CCR / SCR / liquidation penalties / redemption fee) are set at deploy time and immutable thereafter.

Every contract built on an ownable base either renounces ownership at deploy or never initializes an owner at all, so the owner is the zero address and owner-gated functions can never be called. Deploy-time wiring is frozen permanently.

There is no admin path that can:

  • Mint RD outside the borrowing protocol.
  • Freeze user assets.
  • Shut down a branch without trigger conditions being met.
  • Change ratios on an existing branch.

The branch set is fixed at deployment: there is no function for adding a branch to the deployed system.