Curve DAO

Curve’s early pools targeted assets expected to trade near one another. Their invariant bends between constant-sum efficiency near the peg and constant-product protection farther away, reducing ordinary slippage without promising that a broken peg will recover.

Factories later made Curve a family of deployments. StableSwap, crypto pools and newer factory templates carry different implementations, parameters and admin surfaces. “Curve was upgraded” is therefore less precise than naming the affected factory or pool.

LP tokens represent deposited pool positions. CRV does not: it was added to reward liquidity and govern contracts around the exchange, leaving reserve ownership and governance power as separate rights.

The CRV contract began with exactly 1,303,030,303 tokens and a declining annual mining rate. Only the Minter can create emissions within that formula, while GaugeController converts weekly relative weights into each gauge’s share.

Locking CRV for one week to four years creates veCRV. More amount and remaining time mean more voting weight, which decays linearly. Lockers can steer gauges, vote on DAO proposals, boost eligible LP emissions and receive configured fees, but cannot freely transfer the base escrow position.

Curve’s DAO uses separate voting contracts for parameter and ownership decisions. Gauge additions or kills, fee-distributor changes, factory implementation choices and pool administration follow specific contract paths rather than one universal admin key.

Emergency DAO exists for rapid protective actions such as killing gauges or certain pool functions. Speed can limit damage, but also creates concentrated intervention power. It does not mean an emergency multisig can arbitrarily seize every pool’s reserves.

Many pools are deployed as fixed contracts; factories can deploy new implementations without rewriting old instances. Users must inspect the exact pool, factory, owner and kill authority rather than rely on the brand name.

On July 30, 2023, reentrancy locks generated by vulnerable Vyper versions failed in several deployed pools, including pETH-related markets. Attackers could reenter affected contracts. The incident was a compiler-generated guard failure, not proof that every StableSwap invariant was broken.

Recovery and reimbursement proceeded through separate DAO and project actions, including the Curve vest-split tooling and later JPEG’d proposal. The lesson is structural: immutable deployment narrows upgrade discretion but preserves old compiler artifacts, while factory diversity limits some blast radius and complicates risk summaries.