I’ve been in the middle of designing a fully remote, all-discrete integrated amp going on 3 years now :-) So many details o work out, mostly from scratch. switching, switching logic, display logic, EMI immunity, gain stage, RIAA stages with 1500X amplification (can you say "low noise critical?") headphone amps, thermal control, DC sensing and shutdown, discrete volume and balance, human factors, ...... And once something is done, i learn how to improve it. Getting very close to the DFM stage tho, once i fix a few dozen lines of logic control code and refine the active ripple rejection i’m testing out (emphatically not a regulator)
The learning is a ton of fun, and as noted above, the pattern learning of "what makes what difference, and what, surprisingly, makes no difference or harms the perceived sound" (last one is rare, whew)
It has also forced me to go back and do far more extensive benchmarking of my older designs so that i have a reference for improvement. Sometimes that is a flattering experience such as when i measured, under roughly half load, my output stage power supply from 25 years ago, employing tube type PS design to a solid state amp, and measured less than 8mV ripple on the rails. i Concluded that using that product as my test mule was pointless since i was approaching the limits of field measurement. for the non nerds 1mV = 1/1000 volt (rms in this case). In a smaller, vastly cheaper product however the ripple rejection circuitry will yield great benefits. Now to prove/confirm it soudns the part.