This paper examines whether graph-level structural plausibility is sufficient for a curriculum candidate to count as a viable academic architecture under search-based evaluation. Using Campaign 001 of the AlphaPath Evaluation Stack v2, the study analyses the 2005 Civil Engineering curriculum at FACET-UNT against a universe of baseline and generated candidate architectures, with particular attention to phase-wise exclusion across the evaluation pipeline. The paper focuses on a single question: whether institutional translation operates merely as an implementation layer or as a constitutive feasibility filter. The results show that structural plausibility alone is insufficient. All twenty-five evaluated candidates passed the ex ante structural screening stage, whereas attrition emerged entirely at the institutional translation stage, where six candidates were excluded and nineteen survived to the feasible set. The dominant exclusion mechanism was maxₛynchronisationburden, which disqualified two generator families — minimumdurationfeasibleₛearch and lowₛimultaneity — whose graph-level plausibility did not survive translation into term-based academic form. The historical baseline survived all firewalls but ranked fourteenth among the nineteen feasible candidates, showing that survival under institutional translation does not imply competitive architectural quality. The contribution of the paper is therefore not to identify the strongest candidate in final ranking, but to demonstrate that institutional translation functions as a hard feasibility filter in curriculum search. The findings support a methodological distinction between structural plausibility and institutionally viable curricular form, and a further distinction between survival under the feasibility filter and competitiveness within the surviving set. Performance comparisons become meaningful only after both distinctions have been enforced. The baseline’s mid-table position within the feasible set — surviving without being vindicated — illustrates why a curriculum that works on paper is not, by that fact alone, a curriculum worth preserving against algorithmically generated alternatives.
Hugo Roger Paz (Sat,) studied this question.