test(security): cost-basis solvency regression + live front-run fork proof

[drewstone]

Closes the two open questions left after the red-team pass. Both are passing tests.

1. The "HIGH" is provably not exploitable (Regression_Delegation_CostBasisSolvency)

A red-team agent rated DelegationManagerLib.sol:616-694 a HIGH: a delegator holding two same-asset positions to operators with divergent exchange rates supposedly under-realizes a slash loss and drains other delegators. The original exploit PoC failed its own over-withdraw assertion — dep.amount is internally overstated after the slash, but the withdrawable amount stayed correctly bounded.

This regression exercises the full lifecycle the original stopped short of: the attacker fully exits both positions and withdraws everything, then the honest co-delegator withdraws. Two decisive invariants both hold:

• the attacker cannot extract more than their true post-slash entitlement (opX 10 + opY 5 = 15 ETH) — non-vacuity guarded (they really do recover ~15, so the bound is a real test;
• the honest co-delegator is still made whole (30 ETH) — the pool is solvent.

Conclusion: the internal dep.amount overstatement is fully offset and not monetizable. Not a vulnerability.

2. The front-run is live in production (Fork_MigrationClaim_VestingFrontRun_BaseSepolia)

Proves the vesting delegatee front-run (fixed in #190) is real against the actual Base Sepolia deployment, not just the local harness. The test forks the chain, calls the deployed TNTVestingFactory (0x45C1…), and shows a victim's vesting clone delegating its voting power to an attacker on the real ERC20Votes TangleToken (0xB546…). Confirms #190 is needed in prod.

Forking degrades gracefully (vm.skip) when the RPC is unreachable or the deployment is absent, so it never red-fails a sandboxed/offline CI runner. Verified locally against https://sepolia.base.org: PASS, 0 skipped.

Context

Final cleanup from the red-team pass. Fixes: emergencyWithdraw token guard (#189, merged), vesting front-run (#190), subscription terminate gate (#191).

[tangletools]

✅ Auto-approved PR — 5a1e9808

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-25T08:01:12Z}