Modelling Promotion Policies with Survival-Calibrated Agent-Based Simulation: The Promotion Wall and Efficiency-Equity Trade-offs in Engineering Programmes

In prerequisite-heavy engineering programmes, progression rules can redistribute academic risk across time and student groups in ways that are difficult to observe before a reform is implemented. This study uses an agent-based model (ABM) calibrated to longitudinal survival data to examine three progression configurations in a civil engineering degree at a public Latin American university. The historical configuration (Scenario A) allows students to obtain regular course status and defer a separate final examination, thereby accumulating finals debt. Scenario B represents a bundled strict direct-promotion configuration: courses must be fully passed within the attempted term, finals debt is prohibited, and bottleneck-course friction is increased to reflect the stricter assessment environment represented in the calibrated model configuration. Scenario C retains Scenario B and adds a capacity-limited remedial safety net. The model uses 1,343 synthetic agents whose composition matches the empirical cohort observed between 2005 and 2019. The probability of dropout at each model period is governed by calibrated coefficients linking the modelled states of stress and belonging to exit risk. Final estimates aggregate 100 stochastic replications per scenario (402,900 simulated trajectories). Results show six-year dropout of 44.5% in Scenario A, 60.4% in Scenario B, and 53.6% in Scenario C. The low-minus-high-resilience equity gap is 16.2, 26.1, and 20.9 percentage points, respectively. A paired sensitivity analysis using 20 common seeds showed that the comparative ordering of the scenarios was stable under moderate changes in remedial capacity and bottleneck-failure stress. These results are model-implied counterfactual estimates conditional on the specified structural and behavioural assumptions; they are not causal effects recovered directly from observational data. The study illustrates how simulation can extend educational data mining by helping institutions examine plausible efficiency-equity trade-offs before implementing progression reforms.