100 published studies converge on one machine: short daily retrieval practice (she answers with the answer hidden, never re-reads), spaced across days, with immediate elaborated feedback on every item, organized as mastery learning — no new skill until the rung below it is solid — delivered through guided play with a fixed start cue. Confidence is high for that core, moderate for exact session-length and game parameters, and low for mindset messaging and working-memory training as levers.
Her failures are specific, not general — she also scored 100% on some language-arts items and 15/16 on a recent LA test. Each failure maps to a missing rung below it, not to effort or attention. Fix the rung, keep testing it after she gets it right once, and the assignment type stops failing.
Also: feedback nearly doubles the testing effect (g = 0.73 with vs 0.39 without, #4). Only two of ten common techniques rated “high utility” by Dunlosky et al. (2013, #17): practice testing + distributed practice. Donoghue & Hattie (2021, #18) independently rank them top two — distributed d = 0.85, practice testing d = 0.74, larger for weaker students.
If you build only this, you have the evidence base. Everything else is decoration or a weak lever.
Every item is a recall attempt with the answer hidden. Say it, spell it, write it, use it in a sentence. Re-reading and highlighting were rated low utility (#17).
Same minutes split across days beat one sitting. New → Day 1 → Day 3 → Day 7 → Day 21 → monthly. Old lists keep cycling back months later (#15, #16, #22).
Immediate, elaborated, task-level — including on correct answers. Show the answer and the pattern. Never person-level (“you’re smart”) (#4, #44, #88).
No new skill until the prerequisite is demonstrated on a delayed check. Test → correct → retest is worth ~0.5 SD on its own (#48, #49).
Each rule is tied to its studies. Where the evidence doesn’t directly cover 8-year-olds at home, that’s stated — see Limits.
Mastery = 3 correct recalls across 3 different days (#2). A rung passes at ~80–90% on a next-day check before the next rung opens (#48–50). Failed items drop back one rung — never deleted.
No study in this corpus directly tested optimal session length for home quizzing of 2nd graders. This is an inference from the short sessions in the child studies, the spacing literature (#15, #16), and Ericsson’s finding that even elite adults capped sessions at ~80 min (#91). Two 10-minute sessions beat one 20-minute session. End on a success, at a visible finish line — never mid-struggle.
Spacing requires repeated encounters across days (#15, #16, #21). Single-session games show no learning effect (#64). A missed day is fine; a missed week resets the spacing schedule.
Production: say it, spell it, write it, use it in a sentence. Recognition (matching, multiple-choice) only as final format familiarization for school test types (#84, #11). One item on screen at a time; instructions visible, not held in memory (#51–53). Mix skill types within a session, but keep each word list blocked — interleaving words reverses the effect, g = −0.39 (#72).
This is not a skill rung. It’s initiation design:
Guided play — the app sets the target, the child directs the engagement (#65, #66: guided play > direct instruction, maths 0.24, task switching 0.40). Every screen active, engaging, meaningful, socially interactive (#67). Built-in hints, not bare right/wrong (#64). Movement for action words (#96, #97); picture paired with every vocab word, adjacent, no decorative art (#75). Progress shown as a growing collection (words mastered, streaks), not expected tangible rewards — those undermined intrinsic motivation in 128 experiments (#79).
Test the rung before teaching above it. A failure usually means a missing rung below the assignment’s apparent level. Assess rungs 0–3 first — if 0 or 1 is shaky, everything above it is sand. Tap any rung to open it. Amber rungs gate her actual failed assignments.
Direct mapping from the report. No diagnosis path — this is instructional design.
| Assignment | Missing rung | What actually happened | The drill |
|---|---|---|---|
| Vocab color-match F · 3/8 | Rung 7 — word ↔ meaning | Trained by recognition only. Matching exposes her to wrong pairings; without feedback she can learn the wrong match — lure-learning was 24% without feedback vs ~14% with it (#46). | Recall first, both directions: definition → word and word → definition, plus use in a novel sentence. Matching only at the end, always with immediate correction. |
| Commas in a series F · 2/4 | Rung 8 — grammar mechanics | A multi-step rule taught as one step, without a mastery gate. | Drill the two steps separately, then combine: (1) spot the list of 3+ items, (2) place the commas. Mastered = 4/5 novel sentences correct. |
| Spelling tests U3Wk1 12/30 · U3Wk2 16/30 | Rung 4 — spelling patterns | Patterns untaught plus a missing retrieval schedule — words “learned” Monday, gone by Friday (#2, #16). | Pattern-based instruction (formal spelling instruction ES = 0.54; spelling-in-writing 0.94, #42) on the Day 0/1/3/7/21 queue. Never retire after one correct. |
| High-frequency word tests | Rung 3 — orthographic mapping | Partial-alphabetic guessing: she memorizes word shapes / first letters and fails on similar-looking words (#39). | Full mapping per word: see it, sound every letter, spell it, say the meaning, point to the “tricky part” of irregular words (said, they, was, one). 1–4 high-quality exposures can secure a word (#40) — if rungs 0–1 are in place. |
| Phonics tests | Rungs 1–2 or 4–5 | Depends on item type — diagnose by error pattern, not by the test name. | Simple items → check rungs 1–2 (letter–sound, CVC blending). Pattern / long words → check rungs 4–5 (patterns, syllables). |
| Doesn’t start / doesn’t finish → Catch Up folder | Not a rung | An initiation and persistence problem, not a knowledge gap. | If–then plan + ready environment + 2–3 easy wins first + visible finish line. Attribute misses to strategy, not ability (#81, #76, #60). |
The ranked list from the report, verbatim in substance. Study numbers in mono are the corpus entries behind each one.
Condensed one-line findings from the report’s corpus table. Full one-paragraph entries with home implications are in the research notes (01: #1–50, 02: #51–90, 03: #91–100). Search by author, topic, or number. Flags for small samples are carried from the report.
Showing 100 of 100
Honest counting, from the report. Meta-analyses re-pool the same primaries; several entries share one lab or one dataset. Don’t over-count it — and don’t under-count the core either.
Roediger, McDaniel, McDermott, Agarwal, Butler, Kang — 15 entries (#1, 2, 3, 7, 8, 9, 10, 46, 47, 62, 82, 83, 84, 85, 86), ~10% of independent units. Within it, the Columbia (Illinois) middle-school project is one dataset family behind six entries (#7, 8, 9, 10, 62, 86). Karpicke’s Purdue work is a closely allied lab, not fully independent.
Retrieval practice is independently supported by the Purdue, York/UCSD, Erasmus, and Beijing/UCL (Yang) programs and by the Dunlosky/Hattie syntheses — not by WashU alone. ~35 of the 100 entries are meta-analyses / reviews that overlap each other. The novel entries (#91–100) are the thinnest — mostly single studies, for design inspiration, not load-bearing evidence.
Other clusters: Spacing (Cepeda/Pashler/Rohrer) 7 · Dweck/mindset 4 · Sénéchal/home literacy 3 · Gollwitzer if–then 3 · Alloway working memory 3 · Macnamara deliberate practice 3 · Bloom/Kulik mastery 3 · Hattie feedback 3.
The report reports disagreements rather than picking a side. These are the ones that change design decisions.
Kulik & Kulik (1988, #45) favor immediate feedback in classrooms; Butler et al. (2007, 2008) and Mullet et al. (2014, #47, #83) favor delayed feedback for retention/transfer — delayed beat immediate by ~8% on transfer in #47, though 79% of students preferred immediate. Adesope et al. (2017, #5) found the feedback advantage non-significant in their pool. The protocol’s resolution — immediate in-session feedback + delayed re-testing — is a design judgment, not a settled finding.
No study directly tested optimal session length for home quizzing of a 2nd grader. Child-specific studies (Goossens, Vlach, Sobel) are small (N = 36–244). The 10–15 minute recommendation is an inference. The report suggests treating the first month as a pilot: track completion, retention at 1 week, and school-test scores. No study tested a parent-built app of this exact design — the components are each evidence-based; the assembly itself is untested.
Sénéchal & Young (2008, #24): parent tutoring of specific skills, d ≈ 0.51–0.65 — the largest effects. Van Steensel et al. (2011, #25): generic family-literacy programs, d = 0.18, no significant moderators. The gap is attributed to broader definitions and stricter estimation. Design implication stands either way: build tutoring, not a generic literacy program, and expect modest average effects.
Mindset messaging: mindset–achievement r = 0.10; interventions d = 0.08 (Sisk et al. 2018, #77). The national RCT (#78, 12,000+ 9th graders) found GPA gains concentrated in lower achievers — in teenagers, not 8-year-olds. Working-memory training: near transfer only, no far transfer (#54; 87 publications in #55). Design for WM limits instead: one item at a time, instructions visible, short steps (#51–53). Bloom’s 2 SD: from two small dissertations under ideal conditions (FLAG, #48); VanLehn (2011, #100) puts tutoring at d = 0.79. Aslan & Bäuml (2016, #13): one-lab boundary finding — testing potentiated new learning in older but not younger elementary children. It tempers, but does not overturn, retrieval for consolidating studied material at age 8.