iktok90-design/ai-smart-contract-auditor/oracle-redundancy

Oracle redundancy detection

When this applies

Trigger on any of:

• latestRoundData() / latestAnswer() reads from Chainlink or any push feed
• Custom oracle aggregators with primary + fallback sources
• L2 deployments reading any price feed (Arbitrum, Optimism, Base, etc.)
• Rate/price reads that gate borrowing, liquidation, minting, or redemption
• Any "if primary fails, use secondary" branching

Detection patterns

Missing staleness / heartbeat check (HIGH)

(, int256 price,,,) = feed.latestRoundData();   // ignores updatedAt & answeredInRound
require(price > 0);
return uint256(price);

Signal: no updatedAt freshness gate. A frozen feed (feed deprecated, node outage, or a market-wide halt) returns the last value forever; protocol prices off stale data. Require block.timestamp - updatedAt <= heartbeat and answeredInRound >= roundId.

Single point of failure (HIGH)

One feed, no fallback, no degradation plan. If that feed is deprecated or returns garbage, the protocol is bricked or mispriced. Signal: a hard dependency on exactly one external address with no alternative path.

Fallback that silently returns 0 or last price (HIGH)

try feed.latestRoundData() returns (...) { ... }
catch { return lastGoodPrice; }   // ← stale, or worse:
catch { return 0; }               // ← 0 collateral value = mass liquidation, or free mint

Signal: the catch path masks failure. Returning 0 can make collateral worthless (mass liquidation) or debt free; returning a stale cached price keeps trading on bad data. Failures must surface, not be swallowed.

Missing L2 sequencer-uptime feed (HIGH on L2)

// Arbitrum/Optimism: no check of the sequencer uptime feed
(, int256 answer, uint256 startedAt,,) = sequencerUptime.latestRoundData();
require(answer == 0 && block.timestamp - startedAt > GRACE_PERIOD);

Signal: on an L2, if the sequencer was down, Chainlink prices are stale on resume. Without the uptime feed + grace period, the first post-downtime block liquidates everyone at stale prices. Chainlink documents this exact guard.

"All oracles down" path undefined (MEDIUM)

The aggregator handles primary-down but has no defined behavior when every source is stale/reverting — it either reverts (DoS) or falls through to an uninitialized default. Define an explicit safe-halt (pause) rather than trade on uncertainty.

No deviation / circuit breaker (MEDIUM)

Accepting any reported price without comparing primary vs secondary deviation, so a single compromised/erroneous feed is trusted outright.

Severity rubric

Remediation patterns

1. Validate every read: updatedAt freshness vs a per-feed heartbeat, answeredInRound >= roundId, price > 0, and a sane min/max bound.
2. L2 sequencer-uptime feed + GRACE_PERIOD before trusting prices post-downtime (Chainlink reference pattern).
3. Real redundancy: primary + independent secondary (e.g. Chainlink + Uniswap V3 TWAP / Pyth) with a deviation threshold; disagreement → pause, not pick-one.
4. Fail closed: when all sources are stale/invalid, pause the affected operations rather than returning 0, last price, or reverting unboundedly.
5. Circuit breaker on max per-block deviation to absorb wicks.

False-positive notes

• A feed read that already checks updatedAt, answeredInRound, and bounds is fine — don't re-flag for "no fallback" if it correctly pauses on staleness.
• Mainnet-only contracts don't need the sequencer feed — don't flag its absence off-L2.
• A deliberate revert-on-stale (fail-closed) is acceptable when paired with a pause path; distinguish from an unbounded DoS.

Related

• [[oracle-manipulation]] — manipulation is the price-correctness sibling to redundancy
• [[liquidation-cascade]] — stale/zero prices trigger mass liquidation
• [[dos-vectors]] — revert-on-all-down can become a denial of service