So this is what tangent space normals are meant to look like. It's basically a fit function put on normalized normals. The caveat to using this technique is that the camera must be on the correct axis for rendering, in this case lined up along the Y axis. It is then a setting the fit function's source min to (1, -1, 1) and source max to (-1, 1, -1) which correspond to the original normal values. I'm so glad that my mentor figured this out.