Who Registers the Wall? Default rules for delimiting the coalition of observers, registering wall types and arranging the power to trigger

Abstract Where a domain has no screening length, so that the effect of one deployment reaches the whole globe at once, the criterion has a global form: the completion time acting on everyone is that of the fastest of all deployers, and if the coalition of observers must act unanimously, its detection threshold is that of the party least willing to admit error. The fastest deployer and the slowest admitter of error together decide whether everyone has crossed. This paper argues that the structure makes two registrations, the delimitation of the coalition and the classification of walls, key points of governance, and gives default rules for them. First, the trigger arrangement decides the aggregation operator: unanimity takes the maximum, "any party may trigger, all respond as agreed in advance" takes the minimum; with exponentially distributed thresholds, the ratio of their expected detection times is J·H_N, about 8 for four parties and about 29 for ten; this is tier A within that assumption, but under the series' own calibration of thresholds the factor for ten parties is only about 1.3 to 3, the direction is unchanged, and the price is a pause false-alarm rate roughly J times that of a single party. Second, the coalition consists of the parties that bear the consequences, same-source members count once, both in computing distinguishability and in counting the centres able to trigger, and triggering a pause is permissive while resumption is strict, though every pause lapses at the end of its review cycle unless renewed on new evidence. Third, wall type is registered by whether recovery is possible after crossing, not by which safeguards are installed; where boundaries are blurred the default is a strong wall, the party claiming a weak wall bears the burden of proof, and registration and review must be done by a party not sharing a source with the registered system. Fourth, crossing must be distinguished from reclassification: the latter requires no crossing by the system, yet makes a strong wall fall just the same, and is harder to detect; both this failure and secret acceleration are cases where passivity does not hold, and on such cells neither trigger arrangements nor wall registration suffice, the fallback being permission before completion. Fifth, facing incentives to cooperate in words and accelerate in secret, the paper reads the problem as a weakest-link public good and proposes four mechanisms that change payoffs: verifiability, folding spillovers back onto deployers, abandoning unanimity, and constraining only strong walls. Sixth, keeping permission before completion needs not more approvals but removal costs that rise with the speed and scale removal buys. Four testable corollaries are carried over from the book, together with three overturn conditions.