Answers are grounded in this paper and cite their sources.
The paper analyzes how blame affects engagement through various protocols, indicating implications for institutional behavior.
An institution that cannot observe effort can still choose how blame is attached to failure: how much attaches unconditionally, and how strongly the rest tracks what an audience infers. Against a monitor who bears a stake when blame attaches, raising that stake is the obvious lever, and the obvious expectation is that engagement survives up to some level of it and not beyond. That expectation is wrong in a way the instrument itself decides. There are three answers, not two: a protocol may make a rising stake destroy the monitor’s engaged effort permanently, or destroy it over a bounded interval of stakes and hand it back above them, or never destroy it at all. Which of the three a protocol has chosen is settled by two boundaries in the plane of protocols, and the boundaries are objects of different kinds. The first is a ray through the origin. It is exact, it constrains only the proportion between the two channels of blame and not their size, and it is fixed by the audience’s attribution rule together with the technology of detection — neither what the monitor privately values, nor how productively he works, nor the penalty he bears when blame does not attach will move it. The second is a curve rising from the null protocol at the effort private value alone would sustain. It constrains intensity as well as proportion, so the same balance operated harder can lie on its far side; and its position moves with what the monitor privately values, which is the one thing here the institution does not choose. The interval of stakes on which engagement is absent is shown to be an interval at all by an argument that counts nothing: the monitor’s marginal payoff is affine in the stake, so its maximum over efforts is convex, and the region where that maximum is negative is a single interval with at most two ends. Its upper end diverges at the ray as a simple pole whose residue is computed from the primitives. Its closure happens where the recovery manifold’s zero collides with its own pole, which removes both bounding efforts at once and is not a degenerate critical point of the payoff — so the interval’s width vanishes as a square root rather than as the three-halves power a cusp would give. Private value drives the closure down onto the ray at one computable level and reopens it above — on the line of protocols the companion papers report along ; off that line the curve survives that level and so does a bounded interval, at stakes orders of magnitude higher. With no private value at all, the region carrying a bounded interval fills the whole wedge between the two rays this programme has already published. The paper maps and does not rank. What an institution gains by operating inside the interval is a welfare question, and the primitive it would need is not in the substrate.
No takes yet. Share an insight, caveat, or question.
Mikio Hanaeda (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: