i've come to understand practically the problem with too much theory and very little doing

when i began building this new trading infra, i started it with a focus on having the most basic thing i could put into production asap
at the start it was a simple long/short directional model, today is something else
but only now I am working backwards, to the problems that many aspiring traders (a lot on quant forums) try to fix at the start
for example this week my focus was on reducing churn/turnover.
probably something that should be optimized well before putting a model into production right?
well, i tried a bunch of different things, but none of them seem to produce better results than the simplest thing I am doing now
so I tried to get a bit nerdier, and re-think the whole approach into a forecast rather than just pure signals weights
the idea is that instead of just assuming that X signal + a threshold holds enough economic value to churn a position, what if , we assign a number, a forecast of what i expect to get paid for this exposure
that figure is taken from work I've done at the signal layer
i never connected these before, like i said, i just have the simplest thing running
well, it turns out, that the forecasts would be an inefficient way to quantify the signal
for example in the picture below, is a BACKTESTED performance of a model I trade
for this i just ran the forecast model, and assigned to the original backtest, at entry/exit, a forecast based on our prior work
panel 2 shows the distribution of realized return per EV determined by the forecast
one of the bins we made most money is when EV < 0 , which obviously doesn't make sense at all
panel 3 , are the exits by what the forecast said. for example, we'd have lost approximately -1.6k had we not exited positions when the forecast said we should stick to them
or when the forecast said we had no edge, and we'd exited those positions, we'd have missed on $4k profit during that time.
this is MOST PROBABLY my own fault building the forecasts, but that's not the point now. I think all the quants I respect use some sort of forecast to trade based off of, so this is not a method specific take, because if they use it, it's something there
the thing I am saying is that had I started from here, I'd either have a very inferior model to the simplest thing I have running now OR I'd have spent a lot of time spinning my wheels trying to fix this thing rather than monetizing whatever I am trading now OR i'd have skipped on of my best performing trades because I would have thought it didn't work as well as I required it to
make money today type thing
starting with the simplest thing and then working backwards because information is out there in the real world
but now that I have things being monetized fairly nicely, I can focus on the more nerdy things, things that are interesting like this, without the constant worry of opportunity cost on my head