It's especially hard when you have a combination of many government programs at all levels (federal, state, local, many different kinds of taxes, many different kinds of welfare). Even if every individual program uses a gradient, it's still possible that summing all the programs together leads to a >100% effective marginal tax rate.

I think because the gradient is simply too confusing for laypeople to understand.

Even for a simple system like US social security that has a gradient. For every $2 you make over the limit, you lose $1 in benefits. I've heard countless times misconceptions of people thinking they'd be losing money (as in literally having less money net) by working.

I was surprised to see the "Big Beautiful Bill" tax incentive for financing a new US-made car had a gradual phase out above $100,000, although it's implemented as a step function instead of a linear one.

Why do we have legislated cliffs instead of gradients? Because approximately nobody understands lerp. And linear interpolation is the simplest (nontrivial) gradient scheme. Consequently, we get cliffs or, if we're lucky, lookup tables that approximate gradients with stair-step successions of small cliffs.

It's very well established internationally that income taxes are defined by gradients. I have no idea why politicians want to reinvent them so often in other kinds of taxes.

Because that's harder to write the laws for