ADVERTISEMENT

Bitcoin’s 90% Crash on Pyth Was Due to Computers Botching Basic Math

Bitcoin’s 90% Crash on Pyth Was Due to Computers Botching Basic Math

Bitcoin slid about 90% earlier this week on a data network run by several of Wall Street’s biggest players after their computers bungled basic computations.

The Pyth Network briefly reported in error on Monday that the digital currency had plunged to $5,402, far below its level everywhere else.

In a subsequent autopsy of the incident, Pyth said two unidentified firms that supply data to the platform encountered trouble dealing with decimals, causing them to report extremely low Bitcoin prices. Those incorrect numbers got averaged with Bitcoin prices from nine other Pyth contributors, leading to the inaccurate result.

Digital currencies aren’t traded on Pyth. Instead, the service feeds data on the current prices of stocks, currencies (crypto and conventional ones) and metals to software running on the Solana blockchain. Customers use Pyth to, for instance, automatically decide when to sell positions. Pyth said in its blog post that the plunge caused some liquidations.

It was an especially eye-catching error considering the credentials of Pyth’s backers -- about two dozen traders and exchanges that count among the most sophisticated companies in global markets.

One of the contributors used Pyth software that had a bug, according to the blog post. The firm submitted its Bitcoin price as a floating-point number, but the software expected an integer. Instead of presenting an error message, it converted the submitted value to zero and published it.

The other company encountered a race condition, or a situation where software must do two or more things in a certain order to get the right result. The programs took the wrong route, and a zero was used as an exponent when it should’ve been negative 8.

Compounding the problem, Pyth said it gave those erroneous figures too much weight when calculating its aggregated price for Bitcoin.

“Pyth core developers are taking several steps to prevent these issues from happening again,” according to the blog post.

©2021 Bloomberg L.P.