Parsons problems — programming puzzles where the learner is given the lines of a solution in scrambled order and must assemble them into a correct program — have produced a two-decade evidence record showing comparable or superior learning gains to writing-from-scratch on substantially less learner time at the novice level. The field has not, however, adjudicated when in the learner trajectory Parsons problems remain effective and when they should be replaced by writing-from-scratch, a question that becomes acute in the elixresearches program because the program spans four stages from first-construct (Volume 1 of Python Pathway, From-Scratch-ToPython introductory chapters) through consolidation (Volume 2– 3) and algorithmic-fluency (Volume 4–5) to specialisation (Competitive Programming Pathway, chuyên-Tin contest track). The literature splits along a cognitive-load axis: the worked-example tradition and the Parsons-effectiveness camp treat Parsons as a low-extraneous-load substitute for writing at novice level; the expertise-reversal line and contest-pedagogy practice predict that Parsons becomes counterproductive once the learner can generate the lines without effort. This paper synthesises the two traditions and adjudicates the disagreement across four stages and four comparison dimensions: prior-construct-fluency, errordensity-target, instructor-time-budget, and transfer-to-writingtest. The verdict has three parts. First, Parsons problems dominate writing-from-scratch at first-construct and consolidation stages on all four dimensions. Second, writing-fromscratch dominates at the algorithmic-fluency and specialisation stages because the expertise-reversal effect inverts the cognitiveload profile and because contest-grade problem-solving requires the schema-retrieval-and-construction layer that only writing exercises. Third, the boundary between consolidation and fluency is bridged by a Parsons-then-write fade within each lesson — a backward-fading variant where the learner first solves a Parsons version, then a faded-Parsons, then writes-from-scratch. The paper specifies an operational deployment across the four elixresearches series with a per-stage recommendation matrix and a selection rule for when each modality binds. Per the program no-cohort scope, the deployment is described and theoretically validated; no learner data is collected.
That Le (Tue,) studied this question.