If your system can't work without these manual interventions then I am skeptical that it works at all. If these "high level flags" are so important, then it's nonsensical to say they are only 2% of your system.
If they are truly only 2% then just go without them. Go 100% full auto. This would solve all your problems of being exhausted, not being able to make a track record etc.
Look at it like Corporate software company.
You will see developers writing code, software architects, project managers, sales people etc. all foot soldiers doing most of the work.
Then there is middle level management. Overseeing the Foot soldiers and keeping them in channels and intervening when necessary.
Then there is high level Executive management. Making core decisions on Hiring vs firing. Which projects to allocate resources, setting Visions , long term goals etc. etc.
It structured like a pyramid. Very less people at top and +80% of resources should be foot soldiers. We always think that bulk of work is done by Developers. why do we even need Executive management? They are just suckers.
But that is is not how things works. There is value of each layer. Corporate America had done all trials and errors and still more or less follows similar model. It is proven.
Just like that, in my fully automated trading model, bulk of the lower and middle level work is fully automated. System does not need intervention for individual stock entry price or exit price or position size. But High level awareness such as Going how much Long vs Short, OR how much leverage to use? OR minor twists (ongoing) to keep the charts "not to skewed" compare to structure of Market etc etc.
I did started this as goal of 100% fully automation. But in my journey i have realized that there are few things AI is not as good as Humans. There is reason why way over 90% Trading funds are driven my Human Fundamental analysis and stock picking just handful stocks and holding it longer time.
Is it possible to even do that 2% to fully automation? Answer could be yes. But it will need phenomenal amount of coding, news analysis, sentiment analysis, major Events inclusion in database(e.g. Treasuries meetings), scaling in and out modeling etc etc. I Understand what need to be done, but the scope of work is tremendous and will need Team. and even after doing all that it will be debatable (unprovable) to say that which approach is better.
Summary : AI is excellent in crunching big numbers. But for some simple task such as folding laundry, it is pathetic to look at the AI struggle.
I am sorry, I tried by best to explain this here. But I am sure it is not enough. Suffice to say, i have my reasons to keep 2% manual aspects.
