Trying to plot surface using "surfir" - but getting the wrong boundary (due to interpolation error)

I am trying to take the output of a solution to a problem in COMSOL, and construct a smooth surface using the (x,y,z) values. I have come across the very helpful function surfir, which very nearly does exactly what I want.

I have written my own function called mphplot2surf as below, which uses LiveLink with COMSOL to load a model (called "model") in MATLAB and examine the data produced, referring to a plot group known as "pg".

function out = mphplot2surf(model,pg)
    comsolPlot = mphplot(model,pg)
    xCoords = double((comsolPlot{2}{1}.p(1,:))');
    yCoords = double((comsolPlot{2}{1}.p(2,:))');
    zCoords = double(comsolPlot{2}{1});   
    surfir(xCoords,yCoords,zCoords,0.2)
end

and with a particular model and plot group in mind, I get the following plot, where I have added some black lines to indicate where the true boundary of the domain is. It is a T-shaped domain, but MATLAB has incorrectly interpolated between two points on each side and therefore not correctly identified the true boundary. Since surfir relies on interpolation, I can't simply select points on these triangular portions and delete them.

I am stumped as to how to fix this. Could anybody please assist (even if you haven't used COMSOL before)

[Stephen23]

"but MATLAB has incorrectly interpolated between two points on each side and therefore not correctly identified the true boundary."

Actually it is the third-party function SURFIR() that does that. The function SURFIR() relies on

https://www.mathworks.com/help/matlab/ref/boundary.html

to detect the boundary points, which according to its documentation uses an alpha shape:

https://en.wikipedia.org/wiki/Alpha_shape

The "tightness" of its fit can be controlled by its "shrink factor" input argument (default value = 0.5). But it is the nature of an alpha shape (moving a circle/sphere around the points cloud) that it cannot really fit into a arbitrarily tight corners like your concave right angles. It rather depends on the density of your data points, but I suspect that an alpha shape will not work for you.

Why not just interpolate over the whole rectangle, and then mask the points using NaN when plotting?