Full review · 2026-09-30
Review No. 4
30 Sep 2026. What was examined, how, what was left out, and what the review found.
Disclosed
Scope
What was reviewed
- Commit
- Commit 99b07596382536430815378b783a8835adf10fbcActive production implementations and service trust boundaries were reviewed at finalized Base block 51989742.
- Deployed state
- Base (chain 8453), pinned at block 51,989,742
- Toolchain
- forge 1.7.1 · solc 0.8.24 · via_ir, EVM shanghai
- Components
- Production deployment and bytecode correspondence · Vault, collateral strategy and QEURO · Execution pricing and hedging synchronization · Staking, yield distribution and factory registry · Oracle routing and market-price publication · Privileges, upgrades and migration authority · Production service trust boundaries
Method
How
- AI models and tools
- GPT-6 (OpenAI, AI model) — Primary review with three parallel specialist passes · Codex Security (OpenAI, review tool) — Independent repository security scan
- Method version
- v2
- Threat-model-led source review of active production paths, with confirmed, rejected and unresolved hypotheses retained in the private record.
- Deployment verification at pinned finalized blocks, including implementation slots, roles, risk settings and a final drift comparison.
- Runtime bytecode reproduction for active implementations and linked libraries using recorded compiler settings and immutable values.
- Foundry baseline, invariant and pinned-fork tests, including focused proof-of-concept tests for validated findings.
- Static analysis of critical contracts and isolated tests of production services that can write protocol state.
Limits
Not covered
- The review was partial at whole-repository file level and prioritized the active production-critical dependency graph.
- External MetaMorpho governance and implementation, Hyperliquid account custody and the remote watchdog host were outside the available source.
- The first production harvest after the current vault activation had not occurred by the final snapshot.
- Whole-project Slither analysis was limited by tool compatibility; targeted critical-contract analysis completed.
- Exact source correspondence for the inactive Stork oracle remains unresolved.
Deployed code
Is what is live what was reviewed?
Each row compares the implementation reviewed in this cycle with the one live today.
Results
Findings
| Severity | Found | Resolved |
|---|---|---|
| Critical | 0 | 0 |
| High | 0 | 0 |
| Medium | 1 | 1 |
| Low | 2 | 2 |
| Informational | 0 | 0 |
| Ref | Severity | Summary | Resolution |
|---|---|---|---|
| QSR-2026-09-22 | Medium | ExecutionPricing / QuantillonVault. Degraded redemption fallback is bounded by a dedicated exposure budget. | Fixed in ExecutionPricing 1.3.3 · Commit 31ed3d8c962cb89fd54d7551659b469824e42499 · live 1 Oct 2026A dedicated degraded-settlement budget now bounds fallback exposure. |
| QSR-2026-09-23 | Low | stQEUROFactory. Vault name uniqueness is based on the full supplied name. | Fixed in stQEUROFactory 1.0.5 · Commit 31ed3d8c962cb89fd54d7551659b469824e42499 · live 1 Oct 2026Name uniqueness now hashes the supplied content and preserves the existing registered name. |
| QSR-2026-09-24 | Low | UserPool. Administrative token recovery preserves assets reserved for recorded UserPool claims. | Fixed in UserPool 1.0.7 · Commit 31ed3d8c962cb89fd54d7551659b469824e42499 · live 1 Oct 2026Recovery now preserves recorded liabilities for configured assets. |
