feat(proofs): FaultDisputeGame refund mode

protocol/proofs/fdg-refund-mode.md

@@ -0,0 +1,95 @@
+# FaultDisputeGame Refund Mode
+
+## Context
+
+Dealing with `FaultDisputeGame` bonds appropriately is an important part of the current OP Stack
+incident response process. `FaultDisputeGame` stores player bonds inside of the `DelayedWETH`
+contract which imposes a 7 day delay on bond withdrawals. During this delay period, the 2/2 Safe
+managed jointly by the Security Council and the Optimism Foundation can hold bonds from specific
+games or from all games at the same time. Held bonds are transferred to this 2/2 Safe where they
+can be manually disbursed to the correct recipients.
+
+## Problem Statement
+
+Although `DelayedWETH` is an effective tool at preventing a buggy or malicious `FaultDisputeGame`
+from distributing bonds to a potential attacker, the actual process of returning those bonds back
+to the correct players is underspecified. Underspecified proceses can introduce significant stress
+to already stressful situations where executive decisions around fallbacks must be made quickly and
+confidently. We therefore want to introduce alternative low-stress options for bond recovery.
+
+## Proposed Solution
+
+We propose modifying the `FaultDisputeGame` to enter a "refund mode" if the game is blacklisted.
+When the `FaultDisputeGame` is in this refund mode it will distribute bonds back to the users who
+originally deposited those bonds.
+
+### Preventing Complex States
+
+`FaultDisputeGame` currently allows users to begin withdrawing bonds before the game has fully
+resolved. It would likely be necessary to modify the `FaultDisputeGame` to block withdrawals until
+full resolution so that we can more easily reason about the security properties of this change. It
+may otherwise be possible to be caught in strange states.
+
+For instance, `FaultDisputeGame` instances are usually expected to be blacklisted only after they
+resolve incorrectly, so it may be possible for some users to begin withdrawing *before* a game is
+blacklisted, then the game becomes blacklisted and those same users can withdraw their original
+bonds. We can avoid this sort of double-counting and generally simplify the security model of the
+contract by preventing bond withdrawals until the game has resolved.
+
+We also likely need to prevent `FaultDisputeGame` contracts from entering refund mode if any
+withdrawals have already started to avoid dealing with complex state. If any user is able to
+withdraw before refund mode is activated then we assume that all withdrawals could've been
+completed and there's no reason to activate refund mode for those contracts.
+
+### Enabling Blanket Refunds
+
+Existing blacklisting functionality targets specific `FaultDisputeGame` instances and cannot easily
+handle cases where many games need to be blacklisted. If we were to adopt DD#76 alongside
+[DD#77](https://github.com/ethereum-optimism/design-docs/pull/77) then we could use the proposed
+`gameCreationValidFrom` timestamp to act as an implicit blacklist of all games created before this
+timestamp. The `AnchorStateRegistry.isValidGame(...)` function would return `false` if the game is
+explicitly blacklisted OR if the game was created before this validity timestamp. All games created
+before `gameCreationValidFrom` would then enter refund mode.
+
+### FaultDisputeGame Bond Unlock Modification
+
+The `FaultDisputeGame` contract currently calls `DelayedWETH.unlock(...)` when it credits a user
+with a bond. However, `DelayedWETH.unlock(...)` is responsible for (1) starting the 7 day
+withdrawal period and (2) restricting how `FaultDisputeGame` can withdraw out of `DelayedWETH`. We
+would need to move this unlock to only be possible after the `FaultDisputeGame` has resolved.
+Additionally, this unlock would need to be re-triggered if refund mode is activated.
+
+## Risks & Uncertainties
+
+### Security
+
+`FaultDisputeGame` is, obviously, an important part of the security of the current OP Stack. Any
+modifications to the `FaultDisputeGame` contract will likely need to be audited.
+
+### AnchorStateRegistry modifications
+
+[Design Doc #76](https://github.com/ethereum-optimism/design-docs/pull/76) proposes to shift the
+state validation logic currently found within the `OptimismPortal` contract into the
+`AnchorStateRegistry`. Adoption of DD#76 would significantly simplify the specification for this
+proposal. One potential risk here is that this proposal effectively becomes a dependency of
+DD#76 as it would likely be too messy if `FaultDisputeGame` needed to reference `OptimismPortal`.
+
+### DelayedWETH Interactions
+
+We believe that the changes described here would not impact the continued ability to use
+`DelayedWETH` to act as a more serious fallback in the case that refund mode would not properly
+refund users for some reason. We should double check to confirm that we would not see any
+unexpected interactions between the two systems.
+
+### Challenger and Dispute Monitor Changes
+
+Changes that need to be made to the `FaultDisputeGame` bond unlock logic may require some small
+modifications to `op-challenger` and `op-dispute-mon`. We may be able to avoid changes to
+`op-challenger` if we move the `unlock` call into `claimCredit` and simply fail if the unlock
+has already been made. Need to confirm here that `op-challenger` will just retry this claim
+function over and over, but if that's the case then this works just fine.
+
+`op-dispute-mon` would need to be modified slightly to account for the fact that (1) the credits
+inside of `DelayedWETH` will only match *after* the game has resolved and (2) the unlocks may
+become mismatched if refund mode is activated. Both situations can be managed but do require some
+changes to `op-dispute-mon`.