The Price of a Harbour Liability Design when Duty and Shame Share a Lever

Abstract

A safe harbour caps the blame a monitor bears, and the internal-adjudicator companion (companion working paper, 2026) showed that this cap and a monitor’s cultivated duty are the two ends of a single lever on the blame share that drives strategic inattention. That analysis took the harbour’s administration for granted. This paper takes it seriously: a harbour must be applied to named cases, and application is publication, so capping within-reach blame publicizes attribution and pulls the companion’s shame lever in the act of pulling its own. On the verified affine family of blame transforms the two pulls conflict, and the conflict has an exact geometry. Below a closed-form salience intensity $\kappa_c = 1 + \psi_0 \alpha_0 / \delta$ every cap still helps; above it, mild caps backfire — they push the trap floor below its no-harbour level — and only caps tighter than a closed-form threshold $\bar{t}(\kappa)$ do better than nothing. Yet no loudness closes the door: the extinction locus is salience-invariant, and a helpful window of caps survives at every $\kappa$, its existence equivalent to the presence of a diffuse fine. The design lesson is all-or-nothing — a harbour must be silent, or committed enough to clear its own noise; there are no good timid caps, but a working cap always exists. Duty pays no such price: the internal lever publicizes nothing, so salience only widens its advantage. All results are at a single calibration on the single-monitor fold; magnitudes are house values and the composition of salience with the harbour is a stated modelling reading.