Yes, you are right.

The Generator constantly modifies a strategy one step at a time.

It chooses a random modification action of the current strategy:

        const oppIndex: number = Helper.randomInteger(0, 100 - 1)

        if      (oppIndex <   5) operations = ComposerOperations.ChangeStopLoss
        else if (oppIndex <  10) operations = ComposerOperations.ChangeTakeProfit
        else if (oppIndex <  20) operations = ComposerOperations.AddOpenFilter
        else if (oppIndex <  50) operations = ComposerOperations.ChangeOpenFilter
        else if (oppIndex <  60) operations = ComposerOperations.RemoveOpenFilter
        else if (oppIndex <  70) operations = ComposerOperations.AddCloseFilter
        else if (oppIndex <  90) operations = ComposerOperations.ChangeCloseFilter
        else if (oppIndex < 100) operations = ComposerOperations.RemoveCloseFilter

Then there is a complex logic for validation and fixing the chosen action:

        switch (operations) {
...
            case ComposerOperations.RemoveOpenFilter:
                if (numEntryInds === 0)
                    return ComposerOperations.AddOpenFilter
                else if (numEntryInds === 1)
                    return ComposerOperations.ChangeOpenFilter
                break
...

Then there is logic for choosing which indicator to add/modify.

Then each indicator has "own generator" for changing its parameters depending on the position in the strategy.

For example, if the program decides to modify Moving Average indicator and it is "Opening Logic Condition" and:
- it is the first indicator slot - it chooses between "single event signals":
   - Indicator changes its direction upwards
   - Indicator changes its direction downwards
   ...
- if it is second and further. It choses between "continues signals":
  - Indicator rises
  - Indicator falls
  ...

Something like that...

Lets say Express Generator is more clever than EA Studio, which is more clever than FSB Pro.

Thank you! Very interesting and complex + smart, I already thought that this is a feature :-) Superb work!

Still, I sometimes get actually completely identical strategies saved in a row. Have a look at this screenshot. One strategy is a little different (normal as you say), but 2 of them are 100% identical strategies.

Yes, all good, the correlation analysis would otherwise filter them if I wouldn't clear the collection after saving, that's clear and is no problem. Just seemed odd that it would calculate/find 2 times the exactly same strategy in a row, with no changes at all. As far as I understood your logic, the second strategy should at least differ a little bit (e.g. with a different exit indicator or slightly different periods) but it should not have 100% the same logic/settings? It would otherwise mean that it wastes one calculation of a backtest for a strategy it already calculated one step ago.

...and if it optimizes as Mr. Popov says, what target criteria is it optimizing for? Net profit? If so, it would be great if we could choose the criteria via the settings.ini (just like we can in EA Studio), because I always optimize for r-squared and don't care about net profit. This should make EG even more efficient for each users need.

It happens like that:
- The generator finds a strategy that passes all Acceptance Criteria. Pushes it to the collection. Then exports and clears the Collections.
- In the next loop, the Generator randomly chooses to add a new indicator. It may happen that with that new indicator the strategy does not pass.
- in the next loop, the Generator randomly decides to remove an indicator and remove the one that was added in the previous loop. Voila! A strategy that passes appears. It exports the same strategy as in the first step.

> It would otherwise mean that it wastes one calculation of a backtest for a strategy it already calculated one step ago.

Yes, it is.

Alternatively, we may store each strategy and check the list before calculating a new one. However, the list becomes huge very fast. Also checking a new "composed" strategy again all the list is not very much faster than the backtest itself.

Please note that Express Generator (as well as FSB Pro and EA Studio) calculates the indicators and keeps a cache of all indicator signals. It doesn't recalculate the indicator in a case as before.

> ...and if it optimizes as Mr. Popov says, what target criteria is it optimizing for? Net profit?

It is easy to add sorting methods for the Collection. I'll add several soon.
When I start working on Optimizer, I'll use add same sorting methods also.