anshul_marele 

If slightly modify djclements idea, we may avoid COUNTA at all.

=LET(
    data,   UNIQUE($I$3:$K$25),
    first,  CHOOSECOLS( data, 1),
    second, CHOOSECOLS( data, 2),
    left,   UNIQUE( first ),
    up,     TOROW( SORT( UNIQUE( second ) ) ),
    VSTACK(
        HSTACK("", up),
        HSTACK( left,
            MMULT(--(TOROW(first)=left ),--(second=up  ))
        )
    )
)

anshul_marele 

Again, COUNTA gives wrong result. From my point of view more correct will be

=MAP(
    C3# & B4#,
    LAMBDA(A,
        IFERROR(
            ROWS(UNIQUE(FILTER(K3:K25, (J3:J25 & I3:I25 = A)))),
            0
        )
    )
)

anshul_marele I agree with SergeiBaklan... COUNTA is incapable of returning 0 when used with the FILTER function. If FILTER does not find a match, it returns #CALC! because empty arrays are not supported. COUNTA only excludes empty (blank) cells in a range, so error values are counted as 1. Even if you tried using an empty string ("") for the [if_empty] argument, it would still be counted as 1.

In your sample file, COUNTA is returning 1 for "Best2", "Best3" and "Best4" when they should all be 0. If you filter the Lookup1 column by "Best", you will see the only value present in the Lookup2 column is 1.

An alternative approach would be to use the UNIQUE function on the lookup range first, then MAP can be used with a simple SUM or SUMPRODUCT equivalent of COUNTIF:

=LET(
    unq, UNIQUE(I3:K25),
    arr, CHOOSECOLS(unq, 1) & "|" & CHOOSECOLS(unq, 2),
    MAP(B4# & "|" & C3#, LAMBDA(val, SUM(--(val = arr))))
)