My assumption is that when you have "Out of sample testing, percent of OOS bars 30%" enabled, that it only looks at the first 70% of the dataset for all operations (including initial optimization). But I'm not sure about this and it would be great to have a developer confirm this.
Here is what the manual says about the Initial Optimization:
Perform an initial optimization is a default option. It makes the generator try to improve the strategy. This can be done in several ways:
by removing the indicator slots which have no significance for the strategy;
by changing the digital parameters of the indicators, bringing them closer to their default values;
by adjusting the Permanent Stop Loss and Take Profit (if the changing is allowed).
The initial optimization aims at making the strategy more orderly without trying to achieve a higher profit. To achieve optimal optimization, use the Strategy Optimizer.
The Generator starts several optimization attempts only when it founds a more profitable strategy. After the optimization finishes, the Generator continues generating new strategies.
Note the section I've bolded. Based on what I'm reading, it seems that perhaps this initial optimization step is not necessarily looking at any data except final statistics.
For example, if you have an unneeded indicator that can be deleted with no change to the final statistics, then perhaps that is the type of improvement it would make. Again, would be great to have a developer shed more light on exactly what is happening with the "perform an initial optimization" feature.
I do personally use initial optimization myself on both new and existing strategies at the start of every Generator run.