Not easy to code, but is this the reason most EAs don't work?

Hi again Andrey,

Here's my big idea: The reason most EAs don't work over all time periods is, while they might have a good combination of complex filters, markets don't behave the same way over time, they exist in several different states.

Some examples of states I've identified (welcome other contributions / official terms):

- Trending (smooth): Price is moving reliably up or down, and has not touched the MA in a long time, the percentage/ratio of red/green candles is very steady in one direction and is a great position for one solid buy or sell trade, to close when price finally touches the MA.

- Trending (aggressive): Price moves up or down strongly, then there is a counter-movement back down to the MA, but then it bounces back, continuing in the same direction, price moves up or down, bouncing off the faster MA on its way up or down. A good way to identify it would be that it doesn't cross a slower MA. A signal when it touches the faster MA put in an order, to close when it touches the slower MA, then resume 'Analysing mode'

- Swinging: Price is going from highs to lows, EAs using CROSS OUT from the RSI recognising a reversal / swing work reliably with this state, price often crosses MA (but doesn't need to)

- Ranging: Price isn't moving very much, very unreliable trading conditions while market consolidates after big movement - good way to identify would be the number of bars previous with highs and lows all within a certain small range

So, the reason why EAs need Martingale / Hedging / often blow up / work great in some years and terribly in others, it's because there's no easy way to filter for all these different market conditions - The individual vs collective signal for open and close is FANTASTIC in CP, but it is still not able to have different 'MODES' to deal with these different 'states'.

Imagine if there was a way to use different filters once CP has identified what the STATE of the asset is. E.g. it identifies 'Ranging' behaviour and resets all states, starts to analyse, based on a specific set of rules or conditions, that then kicks CP into one of the three modes above to look for orders, and the mode is cancelled/reverted/changed when opposing signals from different states are identified.

I think the different 'states' for all the filters/oscillators might not be too difficult to do, however programming in the 'rules' to tell it which state to use must be very difficult / new.

An example for the Trending smooth rules would be: Previous X bars (e.g. 10), have a green/red ratio of > X and price has not touched MA. Close when price touches MA, or cancel / reset mode when green/red ratio drops below X
An example of Ranging would be: Previous X bars highs and lows have not exceeded X range - Do not make any orders / close all orders
Etc. - This way each mode could have their own SL/TP/Lot size etc. should be much more profitable, but, importantly, RELIABLY profitable, across lots of different assets and conditions.

What do you think?