Deliberately shifting your playing style during a session — tightening up after building an aggressive image, opening up after a stretch of showing down only strong hands — so that the reads other players have accumulated stop being accurate.
The value comes from how opponents actually model you. Nobody plays against your true frequencies; they play against the sample they have seen, and that sample is small and recent. A player who has watched you show down three bluffs will start calling wider, and that is precisely the moment your value bets get paid. A player who has seen you fold repeatedly will start bluffing into you, and that is when calling down becomes profitable. Changing gears means arriving at the new style just before they finish adjusting to the old one.
It is not randomness. Each gear is internally consistent — an opponent watching any single stretch sees coherent, sensible poker — and the timing is driven by what has been shown down rather than by a coin flip. Switching at random produces a style that is simply incoherent, which is a different and worse thing.
The same idea explains why fixed-style opponents are poor practice. A player who is always tight-passive can be solved once and beaten forever afterwards; you learn to read one label instead of learning to read behaviour. Against an opponent who changes gears, a read has a shelf life, and the skill being trained is noticing that the evidence has gone stale — which is the skill that actually transfers to a live table.