Oracle Design

Veil reads prices from Pyth Network  legacy push-oracle accounts without pulling in the Pyth SDK. This keeps the binary small and avoids version conflicts with Pinocchio.

Price Account Layout

The program reads these fields from the raw Pyth legacy PriceAccount bytes:

Minimum required data length: 228 bytes.

Validation Pipeline

Every UpdateOraclePrice call passes through five sequential checks:

1. data.len() >= 228          → OracleInvalid if too short
2. magic == 0xa1b2c3d4        → OracleInvalid if wrong
3. atype == 3                 → OracleInvalid if not a price account
4. price > 0                  → OracleInvalid if zero or negative
5. status == 1 (Trading)      → OraclePriceStale if halted
6. conf ≤ price / 50          → OracleConfTooWide if > 2% uncertainty

Oracle Attack Protections

1. Feed Address Anchoring

The first call to UpdateOraclePrice on a pool records the Pyth feed address in pool.pyth_price_feed. All subsequent calls verify the address matches. This prevents an attacker from substituting a different (manipulated) feed.

First call:  pool.pyth_price_feed = accounts[1].address()
Later calls: accounts[1].address() == pool.pyth_price_feed  ← or OraclePriceFeedMismatch

2. Confidence Interval Check

Pyth publishes a confidence interval (agg.conf) alongside every price. A large confidence interval means the market is volatile or the aggregation is uncertain. Veil rejects prices where the confidence interval exceeds 2% of the price:

reject if: conf > price / 50

This check directly defends against flash-loan-driven oracle manipulation. Flash loan attacks often create conditions where Pyth’s aggregation uncertainty widens (the confidence interval grows) before the price itself moves. By rejecting wide-confidence prices, the protocol refuses to act on manipulated data during the attack window.

3. Status Check

Pyth sets agg.status to values other than 1 (Trading) when price feeds are halted, the market is closed, or there is insufficient data. The program rejects any non-Trading status, preventing stale prices from being used to inflate or deflate collateral values.

4. Negative Price Guard

The program rejects agg.price ≤ 0. Pyth can return negative prices for certain derivative assets; these are nonsensical for collateral valuation.

5. No Timestamp Staleness (by design)

Unlike some protocols, Veil does not check the Pyth account’s timestamp field for staleness. Instead it relies on:

• The on-chain Pyth push-oracle being updated by the Pyth network every ~400ms on Solana
• The status field to detect when the feed is effectively halted
• Keepers calling UpdateOraclePrice regularly

This avoids false positives from short Pyth update gaps during high Solana network load.

Price Caching

Veil caches the latest validated price in the LendingPool struct:

pool.pyth_price_feed  ← feed address (anchored)
pool.oracle_price     ← raw aggregate price (i64)
pool.oracle_conf      ← confidence interval (u64)
pool.oracle_expo      ← price exponent (i32, typically negative)

To convert to USD:

const priceUsd = Number(pool.oracle_price) * Math.pow(10, pool.oracle_expo);
// e.g. oracle_price = 16842000000, oracle_expo = -8 → $168.42

Keeper Responsibility

UpdateOraclePrice is permissionless. Any keeper can call it. Protocols integrating with Veil should run a keeper that refreshes oracle prices before any borrow or liquidation check.

// Refresh the oracle before a borrow
const refreshIx = updateOraclePriceIx(pool, pythFeed);
const borrowIx  = borrowIx(user, ...);
 
const tx = new Transaction().add(refreshIx).add(borrowIx);