The initial guess: all missing values (y-coordinates, red) are taken as the mean=70 of their known sum=280 .

The regression of the red points has slope zero; the regression of the complete dataset is the black line and includes all 8 points.

The y-coordinates of the red points were re-estimated based on the slope of the complete model. Then the red and the black regression lines were recomputed

Second iteration: the slopes of the red and the black line begin to agree.

10'th iteration : a fairly well estimate. All vertical errors (distances of the red points to the black line) look already equal, and because their sum is fixed, the sum of their squares is going to be minimal.

That sum is then also the same as if we had the center of the red points 4 times replicated and taken its distance to the black line.

50'th iteration: final solution having at least ten digits correct. The black colored equation is the final solution for the regression function; the slope in the black nd the red formula has become equal. Unfortunately Excel does not document the sum-of-squares of the residues, but we know, it will be optimal here.