What should be discretionary vs systematic?

I wanted to share what I see as working in my own trading when I use discretion and what is not working.

What tends to be working

1. Predicting/forecasting direction primarily on smallest time frames for a few points at most.
2. Predicting and picking good points where the market is likely to turn on a medium term frame.
3. Tape reading when volatility is higher
4. Bracketing entries and picking off entries with good limit entries. I am very good at picking spots that will limit order fill while not going against me.
5. Identifying the themes that can drive the market over extended periods.

What tends to not be working well

1. Determining position sizing on the “fly”. This is particularly true when trading more contracts where I don’t have a good feel what the size of the losses could be. I have had to reload the OneUp a few times because of this. But, this is also true for “big trades” where I’ve tended to position size too small in the past.
2. Good entry prices. I know that some markets tend to have a micro-mean reversion frequency. However, historically I have did very well picking conditions when markets will state-change. I find this more difficult to do in the more recent markets which tend to award the more conservative limit order entries.
3. Re-entering after stopped out. Determining the stop loss level is really critical and is really more of a function of risk with larger stops almost always performing better. Historically, even if I were stopped out often a reload would work. I find these markets less predictable and reloads not working as well. I have found just using a fixed stop loss to work better in the past because I find it takes too much brain power to determine a good stop. So, I am probably going back to fixed or systematic stops.
4. Overloading system trades that have a large edge has historically been my largest source of losses. The system often will end up making money as designed but due to the larger size and smaller stop loss of overloading the result is negative for me.
5. Trading after the open/morning session. I have found it extremely difficult to make money after the open/morning session. I attribute this to a few sources: the low volatility, higher volatility days tend to be unidirectional in these markets (and I tend to trade counter trend unless I pick the direction in the morning), faster market cognition due to possible super computers or other advanced AI, and also internal/psychological bias causing me to tend to shut down for day when I have a small loss versus trade through the loss.
6. Adding to a loser. I rarely add to a loser. However, given the lower volatility markets I’ve tried to experiment with it but have never found it to work for my style. I think I’d have to trade something with much less leverage for it to work.

What this suggest for future graybox development:

• Position sizing should always be determined systematically. However, a confidence setting would be worthwhile. This could be determined by having confidence levels 1,2,3 and the system automatically determines the position size for the given level of confidence based on statistical risk properties, random entries, etc.
• Entries should be managed by graybox while directional bias is determined by my market read.
• I may need to trade in different styles/methodologies. I might be able to give myself more discretion at open, i.e. a graybox style that allows more input but after the early morning trade should probably switch to a more systematic graybox style.
• I think the current markets probably require trading with some different styles. Being able to switch gears from tape reading when the vol is higher to possibly waiting for pullback entries over the open.
• Stop loss should be automatic.
• Cycle frequency in current markets is “odd”. Typically as markets cycle faster they start to trend but in more recent markets these faster cycles lead to state change jumps to retest lows. This has been throwing me off. I know that the new low is more often bought and is the systematic bias.