Engineering Discipline to Governance Economics: A Proportionate Two-Pass Framework for Algorithmic Monitoring in Capital Markets

Abstract

Institutions deploying artificial intelligence in capital markets accumulate monitoring oversight past the point of diminishing accountability. Context-blind drift detectors impose a retraining tax; regulators mandate proportionality without specifying it. This paper asks how proportionality can be borrowed from heavy engineering and built into governance: an auditable stopping rule at the level of the individual alert. The Two-Pass Context-Aware Framework (2CAF) separates statistical detection from economic adjudication; the dynamic threshold incorporates external stress signals through a Paris-Erdogan power-law form bounded by a governance-set ceiling. The argument draws on Marshall’s diminishing returns, the March of Nines from heavy engineering, and DiMaggio and Powell’s institutional isomorphism. A point-in-time benchmark using electricity load data and market volatility data with injected drifts illustrates execution. Against context-blind baselines (n=100 runs), the context-aware adjudication layer reduced the illustrative retraining tax from $539,672 (ADWIN) to $1,512 whilst preserving accountability. The contribution is architectural: false-alarm costs, missed-detection costs, and the regulatory ceiling become explicit, documented, contestable governance decisions. The framework supplies artefacts US SR 11-7, EU DORA, Singapore Veritas, Japan METI, and FSC Korea mandate but do not specify, all without requiring new primary legislation.