how to use ta.correlation() correctly in for_loop context

I am facing a problem with ta.correlation() within two for-loops. The warning provided by the compiler reads :

„The function 'ta.correlation' should be called on each calculation for consistency. It is recommended to extract the call from this scope“

As pointed out „mandatory“ in the documentation, the ta.correaltion() call should be used on every bar for data_consistency, and should NOT be limited by any condition. In my script (as I understand the Pine5 docu), my for_loop cycles through all 5x5 security-combos, and this is (correctly?) done on EVERY bar. Allthough the warning above is provided.

But the plot (1) of the >>ta.correlation() for_loop output stored to matrix_mcc<< shows, that something is wrong with my looping script results.

A direct plot (2) of ta.correlation() using my arrays is proofing, that correct datas are avail in the arrays for the ta.correlation().

//@version=5
indicator(title = "FAQ-HX1", shorttitle ="FAQ-HX1")

// init a few arrays + one matrix
A_hlcc4 = array.new_float(50, 0)
A_close = array.new_float(50, 0)
A_RlTme = array.new_bool(50, false)
A_ticker = array.new_string(50, "-")
A_currency = array.new_string(50, "-")
A_timeopen = array.new_int(50, 1)
A_timeclose = array.new_int(50, 1)

mcc = matrix.new<float>(25,25,na) //larger than needed



// function fill arrays with pulled tuples
f_FillArray(Name,TickerNumber)=> 
    [src_hlcc4, src_close, src_RlTme, src_ticker, src_currency,src_timeopen, src_timeclose] = request.security( Name, "D", [hlcc4,close,barstate.isrealtime,syminfo.ticker,syminfo.currency,time,time_close])
    array.set(A_hlcc4,TickerNumber,src_hlcc4)
    array.set(A_close,TickerNumber,src_close)
    array.set(A_RlTme,TickerNumber,src_RlTme)
    array.set(A_ticker,TickerNumber,src_ticker)
    array.set(A_currency,TickerNumber,src_currency)
    array.set(A_timeopen,TickerNumber,src_timeopen)
    array.set(A_timeclose,TickerNumber,src_timeclose)
//

//run functions above
f_FillArray("NYSE:PFE",         1)
f_FillArray("SIX:ROG",          2)
f_FillArray("SIX:NOVN",         3)
f_FillArray("NYSE:JNJ",         4)
f_FillArray("NYSE:MRK",         5)

float TEMP = 0
float ValueIn = 0
float ValueOut = 0



for counterOut = 1 to 5
    for counterIn = 1 to 5
        ValueIn:=array.get(A_close,counterIn)
        ValueOut:=array.get(A_close,counterOut)
        TEMP:=ta.correlation(ValueIn,ValueOut,5)
        matrix.set(mcc,counterIn,counterOut,TEMP)
//

// (1)
plot(matrix.get(mcc,2,3), color=color.orange)

// (2)
plot( ta.correlation(array.get(A_close,2),array.get(A_close,3),5)  , color=color.black)

I already asked the Tradingview-Pine5-Support-guys for help. The answer here was, I should run ta.correlation() in global context, unfortunately they are not allowed to provide code examples.
Trying to initiate all variables getting in touch with ta.correlation() via >var< or >varip< shows the same (wrong) results + compiler warning.

Maybe someone can help with correct usage off ta.correlation() in for-loops.

Best Regards
Hendrix

You can't use ta.correlation() in local block. It is due to the way pinescript are compiled before execution.
In your case, you can expand your two loops, it is a workaround, not very 'beautiful', but it works :

matrix.set(mcc,1,1,ta.correlation(array.get(A_close,2),array.get(A_close,1),5))
matrix.set(mcc,2,1,ta.correlation(array.get(A_close,2),array.get(A_close,1),5))
matrix.set(mcc,3,1,ta.correlation(array.get(A_close,3),array.get(A_close,1),5))
matrix.set(mcc,4,1,ta.correlation(array.get(A_close,4),array.get(A_close,1),5))
matrix.set(mcc,5,1,ta.correlation(array.get(A_close,5),array.get(A_close,1),5))
matrix.set(mcc,1,2,ta.correlation(array.get(A_close,1),array.get(A_close,2),5))
matrix.set(mcc,2,2,ta.correlation(array.get(A_close,2),array.get(A_close,2),5))
matrix.set(mcc,3,2,ta.correlation(array.get(A_close,3),array.get(A_close,2),5))
matrix.set(mcc,4,2,ta.correlation(array.get(A_close,4),array.get(A_close,2),5))
matrix.set(mcc,5,2,ta.correlation(array.get(A_close,5),array.get(A_close,2),5))
matrix.set(mcc,1,3,ta.correlation(array.get(A_close,1),array.get(A_close,3),5))
matrix.set(mcc,2,3,ta.correlation(array.get(A_close,2),array.get(A_close,3),5))
matrix.set(mcc,3,3,ta.correlation(array.get(A_close,3),array.get(A_close,3),5))
matrix.set(mcc,4,3,ta.correlation(array.get(A_close,4),array.get(A_close,3),5))
matrix.set(mcc,5,3,ta.correlation(array.get(A_close,5),array.get(A_close,3),5))
matrix.set(mcc,1,4,ta.correlation(array.get(A_close,1),array.get(A_close,4),5))
matrix.set(mcc,2,4,ta.correlation(array.get(A_close,2),array.get(A_close,4),5))
matrix.set(mcc,3,4,ta.correlation(array.get(A_close,4),array.get(A_close,4),5))
matrix.set(mcc,4,4,ta.correlation(array.get(A_close,4),array.get(A_close,4),5))
matrix.set(mcc,5,4,ta.correlation(array.get(A_close,5),array.get(A_close,4),5))
matrix.set(mcc,1,5,ta.correlation(array.get(A_close,1),array.get(A_close,5),5))
matrix.set(mcc,2,5,ta.correlation(array.get(A_close,2),array.get(A_close,5),5))
matrix.set(mcc,3,5,ta.correlation(array.get(A_close,3),array.get(A_close,5),5))
matrix.set(mcc,4,5,ta.correlation(array.get(A_close,4),array.get(A_close,5),5))
matrix.set(mcc,5,5,ta.correlation(array.get(A_close,5),array.get(A_close,5),5))