What you're describing matches exactly the nonce replacement mechanism documented in the article. Bots would send an empty transaction right after block finalization to secure a position in the mempool, then once they spotted a victim transaction, they would replace it using the same nonce but with a higher dispatch class.
The separate EVM channel then ensured that transaction would be included before the victim regardless of arrival order. With MEV Shield and the fixes introduced by the Bittensor team, these attack vectors have been mitigated.
What do you mean specifically by latent bots and anti-MEV bots?
There are a large number of MEV bots operating on Bittensor. One notable characteristic is that all of these MEV bots interact through the EVM side. As soon as a block is finalized, they immediately submit an empty transaction to the mempool without specifying an amount or netuid.
When an MEV opportunity is detected in the following mempool, the transaction is finalized using the appropriate UID and amount while preserving its original position in the queue. This behavior is quite puzzling. It appears that current MEV bots are using a strategy in which they enrich or complete a transaction at confirmation time after it has already been placed in the mempool.
More recently, different types of MEV bots have emerged, including one-block MEV bots, latent (or dormant) MEV bots, and anti-MEV bots.
What are your thoughts on this? What mechanisms do you think these MEV bots are using, and how could they be mitigated or countered?