Escaping the Dark Forest

You put your gold in a box and stuck it in the ground in a ranch in the middle of nowhere. No one knows there is gold in a box in the ground so it's safe. But people know that other people but gold in boxes and stick it in the ground.

One day you go to get it so you load up your pickup with gold digging equipment and drive to the ranch. On the way are spotters. They see your truck has gold digging equipment. They see that the road you're going down leads to the ranch. It's obvious what you're going to do.

They load up their faster Ford Ranger and blaze down the road. You can't catch up. They have a faster car. You get there. They have taken your gold.

If you hadn't gone there, the gold was relatively safe. Maybe some day someone happens on it but realistically probably not.

But you went. By looking for it you revealed you were looking and you revealed where you were looking.

It's much simpler than that. (Also, you appear to have a few concepts mixed up. For instance, one doesn't execute smart contracts, but rather transactions. Smart contracts just sit there until someone sends a transaction to one, at which point it executes that transaction.)

What the bot does is that it checks each transaction that is waiting to be executed and simulates sending that transaction itself on a private blockchain forked from the real network. If the simulation results in a profit, it frontruns that transaction -- i.e., it sends the transaction itself for real, but bidding a higher price than the original sender did, so that its transaction will get executed rather than the the original transaction it's copying.

It doesn't need to perform any sort of vulnerability scan; it just mimics other people exploiting arbitrage or vulnerabilities and pays more to get there first.

Similarly, it doesn't need to adjust any destination addresses. It's just looking for arbitrage opportunities or vulnerabilities that will direct ether to the sender. Smart contracts are entirely capable of getting the address of the message sender, and using that as a destination to send ether to. So the bot doesn't need to adjust the transaction data at all, which would be substantially more complicated.

you will pay a higher fee to a trusted miner to process your transaction without sending to mempool.

It's way safer and less complex than you would expect. Most staking contracts are a copy-paste of two basic staking contracts (from synthetix and sushiswap), so it's enough to do a text diff and see what was changed, which is trivial. For more complex contracts that do something more, funds at risk are the best bug bounty there is - in the current environment if something had >$10M for a month and wasn't hacked, it most likely can't be trivially hacked. Bzrx, the single most incompetent defi platform, was hacked just two weeks after a relaunch for $8M - most likely someone was waiting from the start for it to get enough funds to make the hack worthwhile. Almost no hacks happened during the entire yield farming craze.

Key word trivially - some contracts are custodial, so if someone hacked the owners (or they turned out to be scammers) funds could be stolen, which arguably has a reverse Lindy effect in the beginning. Fortunately people are starting to demand at least timelocks and/or multisigs. Another risk is how well liquidations function during a price crash, for protocols that need them.

The current risk premium was and still is absurdly overestimated, but that was a good thing (for me) as without it three or even four digit APYs wouldn't last a day, but thanks to the unwarranted risk premium they lasted about 2 months. During the short peak three weeks ago it was possible to make even ~8% per day (on millions of dollars - good liquidity), completely risk free (trivial staking contracts). The great crypto bullrun of 2020 already happened and few outside of ethereum even noticed.

You will see billions flow into defi on ethereum as others realize the real level of risk too (which guarantees those astronomical returns are never going to return - but even 10% apy on dollars is good in the current environment).

This as well. Immutable bugs.