In `claim()`, the guard uses `claimed[_treasureHash]`, where `_treasureHash` is an immutable state variable that is never initialized to the caller-supplied treasure identifier, while the contract later marks `claimed[treasureHash] = true` using the function argument instead. As a result, the duplicate-claim check and the state update are performed against different keys, which means a previously claimed treasure is not actually blocked from being claimed again with the same valid proof and `treasureHash`. This breaks a core invariant of the protocol described in the README, namely, that each treasure can only be redeemed once, and allows one valid treasure/proof pair to be reused to drain rewards repeatedly until either the `MAX_TREASURES` cap or the contract balance is exhausted.

The issue stems from a mismatch between the circuit and the contract’s economic assumptions: the Solidity contract is configured for `MAX_TREASURES = 10` and only allows the owner to call `withdraw()` once `claimsCount >= MAX_TREASURES`, while the Noir circuit’s baked-in `ALLOWED_TREASURE_HASHES` array does not actually contain ten distinct treasures because one hash is duplicated and another expected hash is missing. As a result, under the intended one-claim-per-treasure design described in the README, there are only nine uniquely claimable treasures even though the system is funded and accounted as if ten rewards can be legitimately redeemed. That creates two linked consequences from the same root cause: first, one treasure is effectively unclaimable because no valid proof can ever be generated for the missing allowed hash, and second, the normal “hunt over” withdrawal path becomes bricked because honest participants can never reach ten legitimate unique claims, leaving the post-hunt fund recovery logic via `withdraw` function permanently unreachable. The owner can still intervene through the emergency path.