The Observation Portfolio: Why Budget Cannot Move the Blind Spot

A single monitor who splits a finite attention budget across tasks does not, in general, cover the tasks where failure is costliest. We show that across such an observation portfolio, coverage is sorted by the composition of responsibility — the reach-contingent blame differential $\Delta$ that fixes how much of a residual failure is attributed to the monitor — and not by the budget. The marginal blame incurred by observing is largest exactly where the stakes are highest, so the highest-stakes task is the most toxic to look at; under a binding budget the monitor abandons it, and no budget recovers it, because a trapped task is incentive-constrained, not budget-constrained. Only redesigning the responsibility regime — lowering $\Delta$, or raising the production response $g = Q'(0)$ — does. The result holds on one side of a regime boundary whose master variable is $g$, unifying it with the productive-observation cascade of the companion programme; the budget-rationing blind spot that arises on the other side sorts by the opposite attribute, so the two are empirically separable. Recurring control deficiencies on high-stakes tasks then read as a portfolio blind spot, not a resourcing failure.