Oversampling is not one switch

Any non-linear process — saturation, hard-knee compression, clipping of any kind — generates frequency content that was not in the input. For saturation that is the entire point; the harmonics are the sound.

The problem is where they land. A 15 kHz input through a stage generating a third harmonic wants to produce 45 kHz. At a 48 kHz sample rate there is nowhere to put it: anything above 24 kHz folds back down and reappears in the audible band at a frequency with no musical relationship to what you played. That is aliasing, and it is the brittle, metallic quality that makes a bad saturator sound bad.

Oversampling gives the harmonics somewhere to go. Run the stage at 4x and there is room up to 96 kHz: generate the harmonics there, filter above the original limit, come back down.

How much room a stage needs depends on how much new content it makes and how high it reaches. Gentle saturation with a smooth transfer curve produces harmonics that fall away quickly, and modest oversampling covers it.

A hard clipper is the opposite. A hard corner in the transfer function generates harmonics that decay very slowly, and mathematically the series does not terminate at all. There is no ratio that removes every alias from a true hard clip; you pick one where the residue is below anything that matters, and you handle the corner itself separately.

Those two stages sit in the same plug-in with requirements that differ by a factor of four or more. One global switch has to be wrong for at least one of them. Set it to what the clipper needs and the saturation stage burns processor for margin it will never use. Set it to what the saturation needs and the clipper aliases.

Saturation runs at 4x with antiderivative anti-aliasing. The ADAA part matters more than the ratio: rather than pushing aliasing up and filtering it, ADAA changes how the non-linearity is evaluated so that far less is produced in the first place. That is why 4x is enough here rather than 16x.

Soft clipping runs at a fixed 16x. Fixed, not user-selectable — this stage generates the widest harmonic spread in the plug-in, and exposing a control that lets someone set it too low would be offering a choice with only one correct answer.

Hard clipping is exact-threshold, on a fixed 8x path, with polyBLAMP transition correction. The corner is handled by correcting the transition itself rather than asking oversampling alone to clean up afterwards. Correcting the discontinuity and then oversampling costs far less than oversampling enough to make an uncorrected corner acceptable.

True-peak detection runs at 16x with 512 taps per phase. That one is not shaping the audio at all; it only has to be right.

Mastering Suite has no oversampling quality control. Each stage runs at a fixed, separately validated factor — saturation 4x, soft clipping 16x, hard clipping 8x with polyBLAMP correction, true-peak detection 16x — and none of them is selectable. For each stage there is one right answer, and offering a wrong one is not a feature.

An OVERSAMPLE setting does survive in the plug-in, and it is a leftover: it is kept so that sessions saved before the per-stage architecture still load, and it no longer affects processing at all. The status bar states the factors that are actually running, which is the number worth reading.

That is the argument, sitting in a control that does nothing. A quality dial is a question handed back to the person who bought the plug-in to avoid having to answer it. Answering it once per stage, and then reporting the answer, is the more useful thing to ship.