Position

Each "not known" below is a statement about this search, not about the world.

Against the existing reviews

Firth 2021 and Brunmair 2019 (core) review interleaving as a laboratory effect; Murray 2025 reviews spacing in mathematics; none reviews the exercise set as a document.
The EEF review counts classroom studies per strategy (Perry 2021): 12 for interleaving, 22 for worked examples. It does not ask how one sheet combines them.
Du 2020 and Ericson 2022 review Parsons problems as an item type; Muldner 2022 places them among incomplete examples.
Cooper 2006 and Bas 2017 review homework as an amount, not as a sequence.
This review adds: the set read as one sequence of blocks, order and support side by side, solutions and copying in the same frame, and adoption counts.

Interleaving the types of a homework raised delayed classroom test scores: d = 0.79 at 30 days (Rohrer 2015), 1.05 at two weeks (Rohrer 2014, secondary), 0.83 in a cluster trial (Rohrer 2020, core).
Spreading a type across several sets beats massing it, in mathematics too, though less inside courses (Murray 2025, secondary; Emeny 2021).
More problems of one type in one session add little at a delay (Rohrer 2006; Gorgievski 2024).
An example before a problem beats problems alone for novices (van Gog 2011, linked); removing worked steps gradually beats jumping from pairs to problems (Renkl 2002; Atkinson 2003; Renkl 2004, core).
Adapting the fading to the learner beats a fixed schedule in a tutor (Salden 2009).
Parsons problems are as effective as writing code and faster, in the studies of one group (Ericson 2017; Ericson 2018; Haynes 2021).
Online homework with hints and teacher reports gave a small gain at state scale (Roschelle 2016).
Copying homework rises toward deadlines and goes with lower exam scores (Palazzo 2010); solution sites grew during remote teaching (Lancaster 2021).

Whether a whole set built from the measured parts beats the blocked textbook set: no study tests the combination.
What share of a set should be interleaved: Rohrer 2015 illustrates a set of four same-type problems then eight interleaved ones, and no study varies the share.
Whether easy-to-hard order inside a set helps, compared with mixed difficulty: no study found.
When to release full solutions: before the attempt, after submission, or never. One two-cohort study (Rystrom 2005) and many opinions (Q/A threads).
Whether hints written on paper (a folded hint before the folded solution) work as tutor hints do: every hint study found is in software.
How much homework, in problems per type, for retention at a month: only Rohrer 2006 (3 against 9) varies it.
Whether interleaving transfers outside mathematics and chemistry in classrooms (Perry 2021: almost only mathematics).

Variation sets are promoted by England's national centre for mathematics teaching (NCETM 2020); this search found no trial of the set design.
Starred or graded difficulty labels are common (MathEd 2014f) and untested.
Answers to odd-numbered exercises only (OpenStax 2026) is a convention with no study found.
Mastery streaks (three correct in a row in ASSISTments skill builders, Roschelle 2016) are set by convention; the trial tested the whole homework system.

Interleaving raised scores in classroom trials, yet 9.7% of textbook problems are interleaved (Rohrer 2020b).
Spacing and interleaving are judged less effective by the students who would use them (Hartwig 2022).
Long blocks of one type persist, though a few problems per type per session gain as much (Rohrer 2006; Rohrer 2020c).

It reads the exercise set as a document with typed blocks, so a program can check its order: the tags of each problem's type and the position of each block.
It names the baseline (the blocked set with an answer key at the end) and how to compare a shape against it (baseline).
Its proposals are graded: measured where a trial tested the part, convention where only practice holds it (shapes and checks).
It does not claim a new result or the state of the art.