What a Panel Can Measure: Windows, Clocks, and the Identication of Strategic Inattention

Abstract

A research programme on strategic inattention in internal control has established, theoretically, that recurring IT-related control failures can arise as the fold dynamics of a no-collusion equilibrium observed through a one-sided visibility filter: the fall into the inattention trap produces a visible failure event, while recovery is silent. This paper asks the question that theory alone cannot settle: what can an observation panel identify about this mechanism? The answer is governed by a single ratio — the commensurability $W/ \tau_g$ of the observation window $W$ with the institutional failure clock $\tau_g$. A synthetic-data laboratory built on one data-generating layer (an affine blame family that hosts duty, conscience erosion, and liability caps as coordinates of the same payoff) delivers three results. First, the composition of an organization — the share of units capable of falling — is unidentified from failure levels alone: the likelihood carries an exact ridge, broken only when the window bends the exponential ($W/ \tau_g = O(1)$). Second, the dynamic reappearance of the failure signature under conscience erosion is detectable at any window, but its attribution is not: a stakes drift too small to notice reproduces the entire count path, and only commensurable windows discriminate. Third, the U-shaped response to liability caps lives below the silent recovery floor, where monotone evaluation designs return confident verdicts whose sign is an artifact of the sampled range. Each failure of commensurability thus produces a specific wrong answer rather than a blank. The paper states the resulting design requirements for empirical work on control failure and enters two methodological rules — a censoring-corrected independence discriminant and a prohibition on existence accounting — of use beyond this programme.