For now every indicator must implements both values and signals calculation logic. That might be a problem if you want to use values to implement your own signals logic because of performance degradation.

So the current proposal is to make an IndicatorResult as a trait which will only be responsible for representing values. The idea behind is that every indicator should have it's own IndicatorResult type. This will help keep indicators values more verbose.
A single struct, that implements IndicatorResult, must contain all the values, that indicator returns. As a bonus there might be different types of values: not only f64 like now.

Then for signals create another trait Signal and make two internal signal types: BinarySignal for signals like {-1; 0; +1} and FloatSignal for signals in range [-1.0; 1.0]. Every single signal must have it's own struct representation.

This approach has next pros:

• Every indicator may produce different type of values: not only f64s like now;
• Every IndicatorResult may have more verbose documentation and explanation;
• You may choose which signals you want to calculate and which not waste performance on;
• More clear differentiation on signal types;
• Custom signal types might be created for your personal indicator.

Cons:
More code must be written to get a signal from indicator: you need to create IndicatorConfig, then initialize IndicatorInstance (like now), then calculate IndicatorResult, then initialize appropriate Signal, then throw IndicatorResult into Signal and finally get your signal.