<p>This essay challenges the assumption that high-stakes decisions in programmatic assessment for learning (PAL) are inherently intractable. We argue that much of their felt difficulty is diagnostically informative: it signals specific, and in principle modifiable, conditions of implementation. Two conditions are frequently under-developed: the narrative synthesis of assessment information, and the anticipation that decisions emerge from a documented trajectory rather than an isolated event. Where both are met, much of the difficulty specific to programmatic decision-making recedes. This is a position rather than a settled fact, and decisions can remain emotionally, relationally, and institutionally heavy even when well designed. We critique a measurement paradigm that treats competence as a single number and advocate a constructivist alternative in which competence is read as a narrative, while engaging rather than dismissing the psychometric tradition. We identify five institutional domains (value proposition, language, expectations, transparency, and integration) whose cultural transformation realises PAL’s potential, while recognising that workload, infrastructure, governance, and faculty development bound what is feasible, as the uneven history of competency-based medical education warns. Generative AI is the essay’s exigence: by making single-performance assessment newly fragile, it exposes the category error we describe and points to programmatic assessment as a structurally appropriate response. We close with five research priorities, from the phenomenology of non-surprise decisions to the assessment of human-AI collaboration.</p>