In my HLSL pixel shader, SV_POSITION seems to have different values to any other semantic I use. I don't understand why this is. Can you please explain it?

For example, I am using a triangle with the following coordinates:

(0.0f, 0.5f)
(0.5f, -0.5f)
(-0.5f, -0.5f)

The w and z values are 0 and 1, respectively.

This is the pixel shader.

struct VS_IN
{
    float4 pos : POSITION;
};

struct PS_IN
{
    float4 pos : SV_POSITION;
    float4 k : LOLIMASEMANTIC;
};

PS_IN VS( VS_IN input )
{
    PS_IN output = (PS_IN)0;
    output.pos = input.pos;
    output.k = input.pos;
    return output;
}

float4 PS( PS_IN input ) : SV_Target
{
    // screenshot 1
    return input.pos;

    // screenshot 2
    return input.k;
}

technique10 Render
{
    pass P0
    {
        SetGeometryShader( 0 );
        SetVertexShader( CompileShader( vs_4_0, VS() ) );
        SetPixelShader( CompileShader( ps_4_0, PS() ) );
    }
}

When I use the first statement (result is first screenshot), the one that uses the SV_POSITION semantic, the result is completely unexpected and is yellow, whereas using any other semantic will produce the expected result. Why is this?