Today (May 5) at around 15:08 UTC there were a lot of consecutive BTCUSDT AggTrade events that had significant (more than $500) price differentials in opposite directions. For example:
AggTradeId | Price | Qty | First Id | Last Id | Price Delta (from prev trade) | Event Time
1197950573 | 36555.0 | 0.041000 | 2187643044 | 2187643044 | -578.10 | 1651763291926
1197950576 | 37118.5 | 0.004000 | 2187643048 | 2187643048 | 568.5 | 1651763291929
I do not log normal events, so event Id 1197950574 and 1197950575 do not appear in the list above, but I do not think that price differentials of this magnitude over 2 or 3 aggTrade events is reasonable if the matching engine is working correctly and events are being placed in the aggTrade stream correctly. I was not looking at the Binance trading view page, so I do not know what spreads were visible there, but I think it is highly unlikely that there was insufficient liquidity to match prices at spreads much smaller than $500.
I raised a similar issue concerning AggTrade historical data, and provided detailed consecutive examples from Oct 1 2021, in the Binance public data issue page (binance-public-data/issues).
Data discrepancies like this are very concerning - they can significantly adversely affect model-based trading.
Please explain how these discrepancies can happen,