diff --git a/src/CPPIVault.sol b/src/CPPIVault.sol
index 36da2d2..2c6f0bb 100644
--- a/src/CPPIVault.sol
+++ b/src/CPPIVault.sol
@@ -54,6 +54,12 @@ contract CPPIVault is ERC20, Ownable {
 
     uint256 internal constant SCHEDULED_SLIPPAGE_BPS = 50;
     uint256 internal constant EMERGENCY_SLIPPAGE_BPS = 150;
+    /// @dev Wider bound used for an emergency de-risk while the oracle is
+    ///      unhealthy (audit H6). A stale feed serves a last-good price that
+    ///      lags a fast fall; the tight 150bps bound would then make the swap
+    ///      unfillable and brick the very defense it must run. Executing at a
+    ///      wider bound during a genuine feed outage beats not de-risking.
+    uint256 internal constant EMERGENCY_DEGRADED_SLIPPAGE_BPS = 1000;
 
     // ---------- async accounting ----------
 
@@ -414,7 +420,14 @@ contract CPPIVault is ERC20, Ownable {
         }
 
         int256 deltaWad = int256(a.targetRisky) - int256(riskyLeg.value());
-        uint256 bound = trigger == RebalancePolicy.Trigger.Emergency ? EMERGENCY_SLIPPAGE_BPS : SCHEDULED_SLIPPAGE_BPS;
+        uint256 bound;
+        if (trigger == RebalancePolicy.Trigger.Emergency) {
+            // relax the bound while the oracle is degraded so a lagging feed
+            // cannot brick the permissionless de-risk (audit H6)
+            bound = _oracleDegraded() ? EMERGENCY_DEGRADED_SLIPPAGE_BPS : EMERGENCY_SLIPPAGE_BPS;
+        } else {
+            bound = SCHEDULED_SLIPPAGE_BPS;
+        }
         executor.executeRebalance(deltaWad, bound);
         controller.recordRebalance(trigger, a.floor, a.targetRisky);
         emit Rebalanced(trigger, deltaWad, a.floor, a.targetRisky);
@@ -451,6 +464,10 @@ contract CPPIVault is ERC20, Ownable {
         if (address(healthSource) != address(0) && !healthSource.healthy()) revert OracleUnhealthy();
     }
 
+    function _oracleDegraded() internal view returns (bool) {
+        return address(healthSource) != address(0) && !healthSource.healthy();
+    }
+
     function _idleWad() internal view returns (uint256) {
         return SafeTransferLib.balanceOf(asset, address(this)) * assetScale;
     }
diff --git a/src/ExecutionModule.sol b/src/ExecutionModule.sol
index 6c77dba..9fbb445 100644
--- a/src/ExecutionModule.sol
+++ b/src/ExecutionModule.sol
@@ -19,8 +19,11 @@ interface IVaultAccounting {
 /// @title ExecutionModule
 /// @notice Routes every value flow between the legs through Uniswap V3 with
 ///         oracle-anchored slippage bounds. Atomic by construction: no async
-///         dependency anywhere, so the permissionless emergency rebalance is
-///         never blocked (spec invariant 5).
+///         dependency anywhere on the emergency path. The vault widens the
+///         emergency bound while the oracle is degraded (audit H6) so a
+///         lagging feed cannot brick the de-risk; a swap can still revert if
+///         no venue can fill within the (widened) bound, which is the market
+///         genuinely gapping past a fair exit, i.e. the >1/m gap case.
 /// @dev Buy-side funding order: the vault's FREE idle first (settled deposit
 ///      cash awaiting allocation; pending-deposit and reserved-payout cash is
 ///      never touched), then the safe leg. Sell proceeds always land in the
diff --git a/test/CrashScenarios.t.sol b/test/CrashScenarios.t.sol
index 3d78995..c5a3b8f 100644
--- a/test/CrashScenarios.t.sol
+++ b/test/CrashScenarios.t.sol
@@ -257,4 +257,46 @@ contract CrashScenariosTest is Test {
         _try(makeAddr("rando"));
         assertGe(vault.shareholderNav() + 1e15, controller.lastFloor());
     }
+
+    // ---------- H6 regression: emergency de-risk survives a stale oracle ----
+
+    function test_h6_emergencyDeRiskNotBrickedByStaleOracle() public {
+        // wire a real OracleHub as the health source and let its feed go stale
+        MockFeed feed = new MockFeed();
+        MockWstRate wr = new MockWstRate();
+        MockPool pool = new MockPool();
+        pool.setRatioWad(1.2e18);
+        OracleHub hub = new OracleHub(address(feed), address(wr), address(pool), true, owner);
+        hub.refresh();
+        vm.prank(owner);
+        vault.setHealthSource(IOracleHealth(address(hub)));
+        vm.warp(block.timestamp + 3 hours); // feed stale -> degraded
+
+        // execution costs 3% on BOTH fee tiers: exceeds the tight 150bps
+        // emergency bound but is inside the 1000bps degraded bound
+        router.setTier(500, 300, false);
+        router.setTier(3000, 300, false);
+
+        // crash: emergency fires; a stranger de-risks permissionlessly
+        prices.setEth(ETH0 * 60 / 100);
+        vm.prank(makeAddr("rando"));
+        vault.rebalance(); // must NOT revert: degraded bound accommodates 3%
+
+        // exposure was actually reduced (de-risk executed), floor defended
+        uint256 nav = vault.shareholderNav();
+        assertLt(riskyLeg.value() * 1e18 / nav, 0.2e18);
+        assertGe(nav + 1e15, controller.lastFloor());
+    }
+
+    function test_h6_tightBoundStillProtectsWhenHealthy() public {
+        // same 3% execution cost but oracle HEALTHY: the tight 150bps bound
+        // must reject the sandwiched fill (MEV protection preserved)
+        router.setTier(500, 300, false);
+        router.setTier(3000, 300, false);
+        vm.warp(block.timestamp + 2 hours);
+        prices.setEth(ETH0 * 60 / 100);
+        vm.prank(makeAddr("rando"));
+        vm.expectRevert(); // both tiers miss the 150bps minOut
+        vault.rebalance();
+    }
 }