IMHO, you should develop a strategy in this order:

1) First get the basic trade logic working (ie, in your case, flipping between trend and counter-trend trading I believe). Do this only with flat lots of 1 (no reinvestment) so you can accurately determine if the logic itself works;

2) Only once you're satisfied the logic cannot be improved, then consider applying equity-curve trading logic. This simply trades your same logic, but allows the system to go between normal and phantom mode. I do believe the equity curve continues to be calculated for both modes as you suggested, essentially tracking success of the logic itself;

3) Apply OptimalF and/or reinvestment logic. This is a further layer on top of equity-curve trading.

So in essence, you would be starting with a strong foundation (good trade logic), then adding flexibility to turn the system on-and-off dynamically (since we know all systems go through cycles of "not working" and possibly never recover). Finally, you are using the brilliance of OptimalF to maximize your gains on those trades that you do take.

---
The purpose of my philosophy question, above, is to discuss how wide these LowPass-filtered equity curve crossovers "should" be. I guess traditionally I am preferring to see crossovers "not too frequent" but at the same time "responsive enough" to be useful. The crossover in this example is like an on-off switch more than any sort of trade signal. So it's a bit of different thinking.

The main benefit I see from equity-curve trading (ie, utilizing phantom trades) would be the ability to know, quickly enough, when your logic stops working... and/or when it falls out of favor and re-emerges later. This happened recently in my manual trading as the markets went into the summer doldrums period. There would be no way to predict this, except that the equity-curve began falling repeatedly with each successive trade.

PS: I can't take credit for any of the genius code, it's all just snippets from the manual. I was reviewing the workshops the other day and discovered that Workshop 5 had been updated with reverse functions. These are fantastic additions to the toolbox.