Recurve Docs
Website GitHub Open app

Getting started

Core concepts

Four ideas explain most of the protocol: optimistic governance, the float, staked review, and burned slashes.

Optimistic governance

Proposals pass by default. An agent that has to wait for a quorum to say yes cannot trade, because by the time the committee answers the edge is gone.

The protection is not an approval gate. It is two things that cost nothing when the agent behaves: depositors can veto inside a window, and watchers can block. Both are checked by the contract at execution, not by anyone's goodwill.

Vote weight is snapshotted

Weight is read one block before the proposal landed. Nobody buys votes after reading the calldata, and nobody who deposits mid-window gets a say on a strategy that was proposed before they arrived.

Weight is ERC20Votes weight, which is zero until a holder delegates. The app and the SDK delegate to the depositor on every deposit for exactly that reason. The threshold is measured against all shares at the snapshot, delegated or not, so undelegated shares make a veto harder, not easier.

The float and the queue

A vault holding a live position cannot honestly price an exit. Marking it to an oracle lets an informed depositor redeem at a stale number and leaves the loss with everyone who stayed.

So: if the idle balance covers you, you leave immediately. If it does not, you join a queue and claim once the balance covers your shares, at the price then. A queued exit shares in the gain and eats its part of the loss, exactly as staying would. The queue reserves nothing — a strategy can still be funded from the float ahead of a queued request — so a queued exit is a promise about price, not about order.

Review backed by stake

A watcher's opinion is worth what it has behind it. Watchers stake $RECURVE to become eligible, cast one final verdict per proposal, and lose the whole stake if a proposal they approved is convicted.

The asymmetry is the mechanism. Blocking costs nothing — there is no reward to forfeit. Approving a drain costs everything. When a watcher is unsure, blocking is correct. Conviction is a call on the registry that only its governor address may make; the current RecurveGovernor has no path to it, so as deployed a conviction is an action of whoever the registry owner points that address at. See Known limitations.

Why slashed stake burns

Slashed stake goes to the 0xdead burn address rather than to the watchers who blocked correctly.

Paying it out sounds fairer and is much worse. If convictions paid, a coordinated majority could approve-then-convict a minority and farm them, and every watcher would weigh the payout from a conviction against the honesty of its own verdict. Burning removes the payoff, so nobody has a reason to engineer a slash.

Nothing is cancelled, only left behind

There is no cancel on a proposal. A vetoed one is dead; one that enough watchers blocked stays Pending on chain forever, because the contract's block is a revert that undoes its own state write. The agent simply moves on and proposes again. Readers of the chain — the app, the indexer, a watcher — have to treat "pending, window closed, never executed" as the abandoned state it is.