Two records are kept apart: most adopted (what textbooks and lessons do) and best evidence (what a study measured). Upper bounds (what has been shown) are kept apart from lower bounds (what has not).
Sub-question
Most adopted
Best evidence
Measured how
Q1 which aids, used, effective
objectives, bold key terms, summary, checks with a key (OpenStax 2026b)
Gurung 2003; Gurung 2004; Weiten 1999
student ratings and reported use against exam scores; correlational
Q2 objectives
objectives at the start of each section or episode (OpenStax 2026b; Carpentries 2026a)
Klauer 1984 (core); Britton 1985
meta-analysis of objectives with texts; reading-time experiment
Q2 alignment
rubric review of alignment (Quality Matters 2026b); backward design in lesson training (Carpentries 2026b)
none controlled; Larkin 2013, Lynch 2020
one programme before and after; a pilot of two against two courses, not significant
Q3 summaries
a section summary at the end (OpenStax 2026b); key points at the end of an episode (Carpentries 2026a)
Hartley 1979; Cook 1981; Mayer 1996; van der Meij 2012
small experiments on recall and transfer
Q3 glossaries
bold terms defined in context, plus a glossary
Yanagisawa 2020 (second-language only)
meta-regression of 42 studies on word learning (secondary)
Q4 refutation
misconception notes in teaching guides; not counted
Schroeder 2022; Danielson 2025
two meta-analyses, between-subjects experiments
Q5 guided notes
instructor handouts; not counted
Larwin 2013
meta-analysis, 12 college studies
Q5 study guides
publisher study guides and instructor concept lists; not counted
Dickson 2005b; Cushen 2019
classroom experiments, opposite results
Q5 crib sheets
allowed in some exams (Academia 2018b)
Dickson 2005 (core); Dickson 2008; Funk 2011
classroom experiments with surprise tests
Q7 explained solutions
student and instructor solutions manuals (OpenStax 2026b)
49 million students all-time; 5.4 million a year; 71% of United States colleges a year
OpenStax 2026c, the owner's page, undated figures, example only
Quality Matters reach
more than 1,500 institutions joined; more than 60,000 people trained
Quality Matters 2026a, the owner's page, undated, example only
Studies comparing open and commercial textbooks
22 comparisons of learning, N = 100,012
Clinton 2019, a meta-analysis (read)
The counts measure reach, not reading: Gurung 2004 found the use of aids unrelated to exam scores.
What the templates share
Part
OpenStax section (OpenStax 2026b)
Carpentries episode (Carpentries 2026a)
Rubric (Quality Matters 2026b)
objectives at the start
yes, measurable by its own account
yes, with questions, required
objectives (standards 2.1, 2.2) head the alignment set
key terms
bold, defined in context
no rule found
no rule found
summary
section summary
key points, required
no rule found
checks
concept check before the exercises
challenges, with solutions nested inside
assessment (3.1) in the alignment set
answers
key at the back; solutions manuals
inside the challenge
no rule found
boxes
named features (profiles, curiosities, technology)
callouts, instructor notes
no rule found
"no rule found" means the page read did not name one.
All three put objectives first; only the Carpentries template makes them required by its tooling, and the page read does not say which rules its lesson checker applies.
Upper bounds (shown)
Refutation text beats expository text on misconceptions, with no moderator found among 26 (Danielson 2025, secondary).
Guided notes beat students' own notes in college, d = 0.78, and the instructor's full notes, d = 0.27 (Larwin 2013, read).
A summary of captions and drawings matched or beat the full text on transfer (Mayer 1996, secondary).
Numbered in-text summaries raised recall, d = 1.57, in one study of 40 (van der Meij 2012, secondary).
Glossed reading taught more words than plain reading in a second language (Yanagisawa 2020, secondary).
Lower bounds (not shown)
No study found removes one pedagogical aid from a real textbook and measures learning.
No controlled trial of constructive alignment or of a course design rubric on learning was found.
No study of an end glossary in a first-language study text was found.
No study of the revision guide as a genre was found.
Durability of the refutation effect is not settled: Guzzetti 2000 reports long-term effects, and Zengilowski 2021 finds the field's tests mostly immediate.
No added benefit of printed explanations in worked solutions over self-explaining was found (Wittwer 2010).