You are looking at shading, not at geometry

The model is fine. The shading is not. Corners look melted away, dark bands run along the edges, a bevel reads as a row of separate facets, and one face is darker than its neighbour for no reason you can see. So you start guessing: Shade Smooth, Shade Flat, Recalculate Normals, add a Weighted Normal, clear the custom normals, try again. Somewhere in there it either fixes itself or it does not, and you never find out which of the six things did it.

ShadeProof reads the shading setup and tells you which setting is doing it. Then it fixes what you tick, and proves the fix with a number.

The number

Shading corners whose normal leans away from the face it sits on. That is what a dark band is, measured rather than guessed: the renderer lights the corner one way while the surface lies another. On a bracket with inset panels, shaded smooth all over, forty-eight of them. After the fix, none. The count is taken on the mesh your renderer actually receives, after your modifiers — counting the raw mesh would be blind to the very thing being fixed, since Smooth by Angle is itself a modifier, and blind to Weighted Normal and to transferred normals, which is exactly where the first version went wrong.

It will not cry wolf on your models

This is the part that took the longest to get right, and it decides whether the number can be trusted at all. The first version tried to judge by how much of the surface looked wrong, and it was measured against fourteen production assets — a drill, a security camera, an alarm clock, chairs, a lamp arm, industrial pipes, stone statues, a tree. Healthy ones scored anywhere from zero to fourteen percent, and the tree scored higher than several genuinely broken hard-surface parts. No threshold on that number can separate them, and one tuned on a default cube would have flagged good work.

What separates them is a fact, not a threshold: every professionally made model has something holding its hard edges — sharp edges, custom normals, a Smooth by Angle modifier, a Weighted Normal. All fourteen do. So that is what the add-on checks, and it stays silent on every one of them. When a mesh has none of it at all, the add-on says exactly that — nothing is holding the hard edges here, this many shading corners lean away from their faces — and adds that on a sculpt this may be precisely what you want.

What it finds

• smoothing with nothing marked sharp — the melted look, with the count of edges it affects
• flat shading on a model that has bevels, where the bevel becomes a row of facets
• custom normals that no longer match the shape, which is what a mesh looks like when the geometry was edited after the normals were baked
• an uneven scale that was never applied, which skews normals in world space
• a negative scale, which turns the whole surface inside out
• a Weighted Normal modifier sitting before Bevel, weighting geometry that does not exist yet
• a Weighted Normal without Keep Sharp, which will quietly erase the hard edges you marked
• sharp edges marked on an object that is shaded flat, where they do nothing at all
• a smoothing angle sitting exactly where the mesh bends, so the pattern is decided by chance and one moved vertex flips it

Each one carries its own tick box. Nothing is done that you did not ask for.

Where it stops

It does not repair geometry. Doubled vertices, degenerate faces, non-manifold edges, faces pointing the wrong way, faces standing in the same place — those cause dark patches too, and they are a different question with a different answer. When ShadeProof sees geometry as the cause it names it and points at it rather than reaching for it. Two repair tools working on one mesh end in an argument about which one broke it.

It also does not edit normals by hand. There are eight tools for that. This one answers what they do not: which setting is producing the look you are staring at.

It does not touch your scene by looking

Reading is reading. The add-on takes a fingerprint of the scale, the modifier stack, the smooth flags, the sharp edges and the custom normals before and after the inspection, compares them, and tells you in plain words if they differ. That would be a bug, and it says so. On a mesh linked from another file it refuses to repair at all and explains why, instead of reporting a success that never happened.

Tested

Sixty-five checks on Blender 4.2, 5.1 and 5.2 — all three green: fifty-one on what the add-on is for, fourteen on data linked from another file. It also went through an adversarial review by a separate agent, which found eight blocking defects — silence on plain hard-surface models, a false alarm on the canonical Bevel plus Weighted Normal setup, repairs reaching objects that had no finding, deliberately flat faces being smoothed, foliage normals being cleared. All are closed, and the way the add-on measures was rewritten because of them. A clean sphere reports nothing. A coarse organic mesh reports nothing and explains why. A hard-surface part shaded smooth reports twenty edges, and after the repair reports none. The same answer comes out at one hundredth scale and at a hundred times scale, because every threshold is an angle or a fraction, never a distance in metres. Every number above came out identical in three separate runs.

---

Whole deliveries, not one file at a time

• Check a folder. Point it at a delivery and every .blend inside is checked, each in its own background Blender. The scene you have open is not touched — not opened over, not saved, not changed. A file that hangs or crashes takes only its own process down; the other thirty-nine still get checked.
• A report you can send. One self-contained page: verdict first, one sentence a lead can read, then the numbers, then every finding with its address. No internet, no fonts to load, no dependencies — it opens in six months on a machine with no network, which is exactly when someone asks what you delivered.
• Runs without a person. A single command from your pipeline writes the report and returns a code: clean, something blocking, or could not be checked. Those are three different codes on purpose — a check that did not run has not passed, and a gate that treats them the same is worse than no gate at all.
• Remembers last time. Save today's numbers as a baseline and a later run tells you what moved, in which direction, and which findings are new. A measurement that went up is not automatically bad: each product says which way is worse for its own numbers.
• One verdict for the whole delivery. If you own more than one of these tools, one run puts every check you have over every file and answers the only question that matters before you send it: can this go out? Each file is opened once and all the checks see the same state of it.