[pcabc]

what you've done is given yourself a "trend filter" so that you wean out those fakeouts. I used the linear regression, which would work in the same way, likely you and i would have similar results if we just used those.
however as you say the devil is indeed in the detail and yes that tricky part of the right entry, stop and exit are not going to be easy.
in fairness however i'm coming to the conclusion that the benefit of an indicator that gave us our stages would if anything only really be useful in finding the entry

I have several motivations. To see if I can do it well, to see if I can get an indicator that helps me not make bad entries, possible use in screening and perhaps help with market breadth estimations. I did wonder whether I could use such an indicator with an index / sector ETFs using the investors method? I'm trying to whittle down the subjectivity, but man-in-the-loop is required as noisy data, or patterns you did not have in mind designing the filter always seem to get through.

I don't think the drop in 2016 is a stage 4, but at the end of the day who cares?
we would have been taken out with our stop loss anyway. so it all becomes pretty moot to be honest trying to worry about a stage 3 or stage 4 exit as we should have taken our profit by then. and the stop loss is going to be nigh on impossible to really get right, so all we're doing it for is the entry

The close was below the falling MA, there are lower highs and lower lows. On this basis definitely stage 4. However, stages 2 and 4 are well defined, whereas 1 and 3 seem more subjective. Since any MA will only be perfectly flat (not rising or falling) using a simple decision based on the position of the close wrt to the MA and its slope will whipsaw around 2 & 4.

now, if i could code a double or triple top breakout where the SMA on the candle chart is sloping up, i'd be an extremely happy man. and my coding skills are just not good enough. you will be the first to hear by the way if i do manage it!!

Yes, how that works is quite tricky to conceptualise from a coding perspective. I have a couple of screeners. One which ranks stocks based on a number of parameters including volume, breakout from channels, MAs and another which detects breakouts from manually entered trendlines. I usually set the trendlines based on the output of the first filter.

I think the core of this would be detecting whether the price today was higher than the upper limit of a channel yesterday / last week. Effectively automatically setting a flat breakout level. Won't work for sloped breakout levels. OTOH I think the flat approach hits many.

I don't think there is a perfect screener. The best we can do is to use several good screeners.