The Pair and the Believer: Attribution, Affineness, and What a Blame Protocol Cannot Buy

An institution that assigns blame after a failure has, on the face of it, two dials: a floor the monitor bears whatever the audience concludes, and a coefficient on what the audience concludes. Companion A (2026) showed that the first is not really a dial of attribution at all — it trades against the monitor’s private production value at the stake, exactly and identically in the action — while the second has no such partner, and it asked whether that asymmetry is a property of attribution or of the affine form the protocol happens to take. It is neither. This paper shows that the asymmetry belongs to the pair formed by what the audience reads and what the detection technology delivers. Replace the audience entirely and the floor’s substitution is untouched, because the floor’s coefficient does not mention the audience. Hold the protocol affine and re-pair it, from either coordinate, and the audience’s coefficient acquires an exact substitute — provided the audience’s share is a particular kind of function of coverage with its pole at one particular place, a place made of the stake and the residual cost. But an audience that reads greater care as greater culpability cannot hold such a share. One identity does it: the audience’s second branch is prevented at least as often as the monitor’s own exactly when its posterior stands above its prior, and the posterior begins at or above the prior whatever the audience believes. So a posterior that does not fall in coverage forces the inversion everywhere — with no appeal to the pole and none to the shape of the branch. The escape is a posterior that falls, an audience for whom failure under high coverage is evidence of bad luck rather than of neglect; the paper declares the monotonicity as a convention and prices it rather than claiming the pair away. The pair exists as algebra, and no audience of the kind this programme uses can hold it. Within a class of attribution rules defined by feasibility rather than by functional form, and at a fixed stake, the substitution set is therefore closed: the floor’s is the whole of it, and adding parameters to an attribution rule buys behavioural reach but never a second way of undoing a change in private value. Releasing the stake opens a further door, and the same door is opened by setting the residual cost to zero or by unbundling the two legs of the payoff — three removals of a single dependence rather than three separate exceptions. Whether a fourth route exists is not settled here.