The suggestion card computes the masking margin for exactly your setup: more voices mask each other (so the bed can sit closer, plateauing around four), a mixed-gender or wide-panned cast stays separable (so it leans deeper), a bed that is thin or scooped where speech lives needs extra room, and a bed that plays with gaps leans deeper still, because words slip through momentary dips. Declaring headphones credits wide panning with a small adjustment. Speaker playback does not receive that adjustment.
On top of those, a small constant allowance covers spoken peaks: the bed is anchored by its average loudness while voices are set by their peaks, so the fader math alone reads the voices a touch quieter than they really are. Measured across the studio's own voices, that gap averages about 2 dB. That is why a suggestion that once felt a tick too loud now sits a little deeper.
The gold tick on the margin control is the computed starting point and the green band is the suggested range. “Reset all levels” restores that starting point in one move. Everything stays advisory: drag any fader yourself at any time.