Unfortunately, most what you wrote is incorrect because you are confusing decoding and filtering by mixing them together.
Let's start with reconstruction filters. You MUST have them if you have any hope of reproducing the original recording. This is a baked-in consequence of digital sampling theory as invented in the middle of the last century by Whittaker, Nyquist and Shannon. It does EXACTLY the same thing whether you have a current SOTA DeltaSigma, an original Phillips 4x oversampler, a ladder or R2R DAC. The filter's job is to allow everything through without any impact for up to half the source sampling rate and then to plunge precipitously down to block everything above half the sampling frequency. Sure, with certain test signals, filters ring, but good ones don't when playing music.
Then let's look at the DAC itself. Assume we have a mathematical system where 1 = the quietest breath of a gentle breeze and where 256 = the loudest bit of an explosion. This can be encoded in an 8-bit digital word. When all the bits are 11111111 = 256; when all the bits are 00000000 =1. The challenge is converting these bits into a voltage which can make speakers move. Suppose we set the first bit flipping on to be 0.01V, when the second bit turns on, it has be be double that at 0.02V. The third bit flipping on is double again - 0.04V. This 2x mathematics with each bit needs to be extraordinarily accurate. The simplest architectures to understand are ladders and R2R DACs where each step is done with precision resistors. It's easy to understand, but spectacularly hard to get right. And if you do get it right once, it will drift with temperature and with time.
The result is - even the very best R2R and Ladder DACs struggle to get close to the near-perfection achieved by DeltaSigma. Where they are at their worst is not at the very quiet or very loud part, but slap bang where the music is. They add distortion. It's probably not possible to hear this distortion, however. BUT this has NOTHING to do with filters.
