Why scaling effect backface while seems not effect winding order?

For example, in a left-hand system, we have a triangle, p0(-1, 0, 1), p1(-1, 0, -1), p2(1, 0, -1) with indices[p0, p1, p2], not sure of the orientation, just choose the one that make triangle’s front face world normal to be (0, 1, 0). Open backfaceCulling, so we can see the triangle using an up-to-down camera.
Then I give the face a scalingY = -1, it seems the winding order is not changed as points will not move after scaling, but the front face world normal changed to (0, -1, 0).
But I learned after searching that the winding order decide the frontface and backface.

face front is determined by the winding : triangles indices. And face culling sets if front or back is rendered.
Normal is part of the attributes (along UV, position,…) and it’s transformed by the mesh matrix. So applying a scalingY = -1 will change normal.
position are transformed too but the result will give the same position.

Thanks for replying but I still dont understand, maybe I dont make myself clear enough.
After scaling the frontface is below and the backface is above. From what I searched, gl judge if a face is front or back by the winding order.
However, triangle indices dont change, points positions dont change and camera dont change, so winding order dont change, they just conflict, thats why I ask the question.

Can you open a PG to illustrate it?

There is also an orientation flip based on matrix determinant : Babylon.js/packages/dev/core/src/Layers/thinEffectLayer.ts at 86bda66b6f61e482374c1a0597f1f504cd75837d · BabylonJS/Babylon.js · GitHub

That’s the reason there is a difference with expectation between mode 0 and 2 in your PG.

Then it works as expected, thanks!