On Monday, 20 September, a Solana-based price data provider used by some Wall Street participants showed an anomalous drop in Bitcoin to $5,402.
Between 12:21 and 12:23 UTC the Pyth BTCUSD aggregate price was below $40,000 – the lowest price reported was $5,402 with a confidence interval of $21,623 (4x the asset reported price) for a single slot – which was off-market relative to the BTC price available on other markets
— Pyth (@PythNetwork) September 20, 2021
The Pyth platform team, whose clients include financial giants Jump Trading Group, DRW and the FTX crypto exchange, said on Twitter:
«Between 12:21 and 12:23 UTC [15:21 and 15:23 MSK] in the BTC/USD pair the aggregated price fell below $40 000 — the lower reported price was $5,402 with a confidence interval of $21 623 (four times the asset-reported price)».
The outage affected only the BTC/USD pair and did not affect other Bitcoin markets, according to Pyth.
In the course of the investigation, engineers of the platform found that the cause was data provided by two projects on the Solana blockchain. Because of a mismatch in the data format, the platform’s software treated the price they quoted as near zero. The Pyth algorithm also gave the data a high weight when calculating aggregated quotes.
According to the Pyth team, the outage affected several Solana protocols. Engineers mainly linked this to their use of prices without accounting for the confidence interval.
For example, the Bonfida project reported that the price anomaly triggered a series of liquidations of Bitcoin perpetual options on its Audaces platform.
Today around UTC 13:22, the @PythNetwork oracle’s BTC/USD price feed suffered a flash crash with the price of $BTC going as low as $8k. This caused a series of liquidation events on the Audaces protocol BTC-PERP market (unfortunately working as intended).
— Bonfida (@bonfida) September 20, 2021
Developers at Pyth said they would implement a number of changes to the software to prevent similar incidents in the future:
- introduce testing during integration of protocols to prevent data-format mismatches;
- improve monitoring tools so data providers can respond more quickly to anomalous prices on the mainnet;
- adjust the logic for weighting the supplied data;
- develop documentation and code samples for integrated Pyth protocols emphasising the joint use of price and the confidence interval as a more accurate market signal.
Earlier, the incident followed a broad outage on the Solana blockchain — new blocks did not appear for about 17 hours. Validators restarted the network on September 15.
According to the project team, the outage was caused by a spam attack by bots during the IDO token sale of the Grape Protocol on the Raydium platform.
Follow ForkLog news on VK.