Public policy increasingly confronts problems whose informational complexity, systemic interdependence and uncertainty exceed the cognitive limits of conventional policy processes. Institutions respond to these conditions through distributed expertise, scientific analysis, administrative knowledge and computational tools, yet the use of computation within policy-making is often framed too narrowly as a means of prediction, optimisation or technical support. Such approaches risk overlooking a broader transformation: computational systems can participate not only in the analysis of predefined alternatives, but in the representation of public problems, the integration of heterogeneous evidence, the exploration of wider policy possibility spaces, the examination of trade-offs and uncertainty, and the recursive redesign of interventions in response to implementation evidence.
This Working Paper develops the Computational Policy Design Framework (CPDF) as a model for understanding these processes within the wider architecture of Institutional Cognition. Building upon the Institutional Cognition Conceptual Architecture and the Institutional Cognitive Architecture Model, CPDF treats computational capability as one component within a socio-technical policy-design system rather than as an autonomous source of policy authority. The framework distinguishes computational exploration from legitimate institutional judgement and examines how policy representations, evidence architectures, constraint architectures, policy possibility spaces, trade-off spaces, uncertainty, scenarios and feedback can be organised into a recursive process of policy learning.
Particular attention is given to the epistemic risks created by computational formalisation. The paper introduces the concept of computational legibility bias to describe the tendency of institutions to privilege forms of knowledge that are easier to encode and process, while potentially marginalising qualitative, contextual or tacit forms of evidence. It also develops the principles of model pluralism, epistemic traceability and Policy Model Provenance in order to preserve the visibility of assumptions, evidential sources and model evolution across policy cycles. These mechanisms become especially important as artificial intelligence, generative systems, computational agents and policy digital twins expand the scale and autonomy of computational participation within governance.
CPDF further reconceptualises implementation as an epistemic phase of policy design. Policy interventions generate evidence concerning the adequacy of prior representations, assumptions and models, allowing implementation feedback to contribute not only to parameter adjustment but to the possible re-representation of the problem itself. This recursive structure supports adaptive policy pathways, staged experimentation and learning before scale, while preserving constitutional and normative boundaries around legitimate decision-making.
The framework therefore positions Computational Policy Design between conventional policy analysis and automated governance. Its principal scientific contribution is to establish computation as a means of expanding the space of institutional reasoning rather than replacing collective judgement, providing a research architecture through which computational methods can be reconstructed, compared, validated and experimentally investigated as components of accountable, adaptive and cognitively augmented policy-making.
