An accurate forecast quietly assumes you won't intervene

A prediction of an outcome is a statement about a system you are standing outside of. The moment you can move the thing you are predicting, the prediction curdles: its accuracy was bought by assuming you would hold still. So the better a forecast describes what happens to you, the more it has quietly cast you as a spectator. For anyone inside the loop, the useful object is not the forecast. It is the seam where the trajectory is still loose enough to take a push.

This is why telling a President a faction has [a 70 percent chance of winning a civil war](Forecasts are useless to any actor large enough to change the outcome) tells him nothing. He is the largest variable in the situation. The odds are downstream of choices he has not made yet, so a number that treats them as fixed is true and inert. The same inertness shows up two feet from a screen. Knowing the name of the manipulation pattern running in a video predicts, accurately, that it will hook you, and that knowledge changes nothing on its own. It does work only when it stops being a forecast and becomes [a perceptual interrupt](Naming a pattern you are inside of creates a pause you can actually act on), the half-second where you close the tab instead of watching the prediction come true.

So position decides what knowledge is worth. Watch from outside and accuracy is the whole game. Step inside and accuracy is just a tidy description of your own passivity, and what you needed was the lever, not the odds.

Source claim: For an agent embedded in a system, predictive accuracy and actionable leverage point in opposite directions, because an accurate forecast presupposes the agent's own non-intervention.