Shader Graph - Is recompilation of a shader for each change a good idea?

I am writing a Shader Graph for my engineering thesis. I'm kinda stuck on whether I should recompile the shader each time some pin value changes to get a real-time preview result or maybe there's a better way to handle this.

Let's consider this screenshot:

For simplicity a fragment material has only 2 fields - base color and alpha.

the generated code of this graph looks like this:

void FragmentShader(out FragmentMaterial material)
{
    vec3 _Vector3_4302458404873090201 = vec3(0.2, 0.0, 0.0);
    material.baseColor.rgb = _Vector3_4302458404873090201;
    material.baseColor.a = 1.0;
}

it's just a simple function that is added to the end of a shader code and gets invoked in a main function of a fragment shader.

User can manipulate X, Y and Z values. Each time the value changes the shader is recompiled and cached. Maximum cache size is 50 shaders.

The number of values ​​that can be used is very large. Let's say the user will be sliding the X from 0 to 1 (not too fast) increasing by 0.01. The value is always different, so that's 100 shader recompiles. Only one shader can be recompiled during a frame.

Although I don't see a problem with that, because the shader compilation is fast, I feel like this is not the best option.

Or maybe I'm just tripping and until it won't cause me a performance problem I should stick with it since it's just an editor. I believe the slowest part would be preprocessing and compiling to spirv (which may be kinda slow, I used to use shaderc library), but it will be done when the final values were picked (if they're not in a uniform buffer, but it does not require a shader recompilation).

The editor uses OpenGL so it does not require a shader to be in a spirv format to work.

I don't know. Do you have any ideas, or maybe do you see some problems with this approach, or could suggest what I should do?

As Quimby suggested, for previews the shader is recompiled only when connections change.

If an input pin is not connected then it's treated as a uniform variable.

If it's connected, then there has to be a different source (node) that probably has tweakable inputs.

Unless it's a property node that is always treated as a uniform variable and my existing workflow already supports it, so I didn't have to change the way the code is generated by these nodes.