CAPIRE Intervention Lab: An Agent-Based Policy Simulation Environment for Curriculum-Constrained Engineering Programmes

TL;DR

The paper introduces CAPIRE Intervention Lab, an agent-based simulation environment that enables testing of curriculum and policy reforms in engineering education to reduce dropout rates and improve student success.

Contribution

It presents a novel agent-based simulation environment that integrates predictive analytics with in silico experimentation for policy testing in engineering programs.

Findings

Policy bundles targeting early courses reduce dropout by ~3%.

Simulation shows increased course passing for vulnerable archetypes.

The environment enables dynamic policy testing before real-world implementation.

Abstract

Engineering programmes in Latin America combine high structural rigidity, intense assessment cultures and persistent socio-economic inequality, producing dropout rates that remain stubbornly high despite increasingly accurate early-warning models. Predictive learning analytics can identify students at risk, but they offer limited guidance on which concrete combinations of policies should be implemented, when, and for whom. This paper presents the CAPIRE Intervention Lab, an agent-based simulation environment designed to complement predictive models with in silico experimentation on curriculum and teaching policies in a Civil Engineering programme. The model is calibrated on 1,343 students from 15 cohorts in a six-year programme with 34 courses and 12 simulated semesters. Agents are initialised from empirically derived trajectory archetypes and embedded in a curriculum graph with structural friction indicators, including backbone completion, blocked credits and distance to graduation. Each agent evolves under combinations of three policy dimensions: (A) curriculum and assessment structure, (B) teaching and academic support, and (C) psychosocial and financial support. A 2x2x2 factorial design with 100 replications per scenario yields over 80,000 simulated trajectories. Results show that policy bundles targeting early backbone courses and blocked credits can reduce long-term dropout by approximately three percentage points and substantially increase the number of courses passed by structurally vulnerable archetypes, while leaving highly regular students almost unaffected. The Intervention Lab thus shifts learning analytics from static prediction towards dynamic policy design, offering institutions a transparent, extensible sandbox to test curriculum and teaching reforms before large-scale implementation.