Q: What am I missing?
MT4-Terminal, a programmable, client-side trading access-device uses ( for historical reasons ) a time-reversed-indexing of registers, that store actual / historical values { open, high, low, close, volume } for respective time-units ( specific in duration for each TimeFRAME ).
Other values get "derived"/calculated from these cardinal values ( either by "pre-defined" ( implemented ) indicators { iRSI(), iMFI(), ... } or programmatically by your MQL4-code ).
The first common issue is, that time passes... ( obvious..., but surprisingly, there are some artificial side-effects in PriceDOMAIN, that surprisingly originate from TimeDOMAIN, introduced by the historical simplifications for the fastest possible handling of pricing data in real-time, that cause troubles alike your idea stroke into ).
Once aNewBarEVENT appears, a very special case happens ( always )...
Registers open[0] == high[0] == low[0] == close[0] & volume[0] == 1
As the time passes on and via subsequent anFxMarketEVENT arrivals into your MT4-Terminal software both volume[0] > 1 grows & register close[0] gets newer and newer value(s) ( typically other than open[0] )
That mechanism thus "complicates" any decision making bound to "just-present" value of close[0], as the next arrival of anFxMarketEVENT ( a new Ask/Bid prices from Broker ) moves the register value to some other value, than might have been used in previous (milli-)second.
This way your code may have started some activity right in the moment ( a derived ) value of iRSI( ..., PRICE_CLOSE, ... ) was indeed > 70, however during the remaining time till anEndOfBAR the subsequent close[0] prices went in such a direction, that the last value of iRSI() upon "leaving" the Bar was way less than 70 ... ( as you seem to try to depict in your ( cropped w/o scales ) PrintScreen )
The second issue seems with aPriceDOMAIN distance of the markers "above" the candle close[aBarPTR] level.
A LONG trade was entered at a price given by both the Bid == close[0] & Spread, nevertheless the graph shows candle prices as Bid-based prices, so in case the visual distance equals to Spread at the time of XTO ( where Spread value might be of a constant or variable size, depending on your Broker Terms & Conditions ), the position of a LONG trade entry is fair & correct.
A third issue that may be but need not be valid in this case is that MT4-Terminal feature called a Strategy Tester was not quite fair & handy in handling a multi-TimeFRAME calculations during back-testing. Some years ago, our team simply gave up to use in-built functions and started to use our own register-handling so as to maintain our own, independent, multi-TimeFRAME values we could really rely on.
Q: How can I fix it?
Due to the nature of handling the hot-[0]-bar events, your code needs some other filter(s) or additional indicators, that do not lag ( do not add additional latency & skew in time, when a trading decision gets triggered ) and do not get flutter ( repeating triggering during the same bar ).
MQL4 gives some tools for doing that, if in a need to get more, one may resort to additional external facilities ( a fully distributed multi-agent grid/cloud/GPU-based computing ).
So, enjoy the great world of FX/MT4.
All these possibilities are at your fingertips.
