Research is a design input, not a marketing adjective.
We keep a working collection of vocabulary, reading, and memory research to guide what we build and what we prioritize. We begin with meta-analyses, systematic reviews, and findings that have held up across studies, then ask a second question: can this result become useful, humane practice for a reader working with a real book?
Choose words worth the reader’s time.
Not every unfamiliar word is equally useful. The app begins with words that actually occur in the selected book, then considers reading difficulty, frequency, contextual diversity, academic value, and productive word parts. The goal is vocabulary that can help beyond a single sentence.
Selection is also subtractive. Already-mastered words, proper names, unsupported forms, and low-value function words are removed before a word can enter a lesson.
Product decision: prioritize transferable, appropriately challenging vocabulary from the book instead of working through every rare word or a fixed universal list.
Keep the word connected to an authentic passage.
Vocabulary instruction is stronger when a learner gets both clear information about a word and repeated opportunities to meet and use it in meaningful context. A sentence from the current book gives the word a reason to matter; an accessible definition makes the learning explicit.
The app checks the exact occurrence, not merely the spelling of the word. It rejects a passage when the sense is wrong, the use is idiomatic, the surrounding clues mislead the reader, or the occurrence is not teachable. If a real source span cannot be recovered, the app skips the word rather than inventing an excerpt.
Spoken pronunciation is part of the lesson, but written IPA is not displayed. Words from an added book have saved pronunciation audio ready before play; the built-in trial can use the browser’s voice instead. Some meaning exercises may also include a generated illustration. Illustrations are optional, and a missing image never prevents the lesson from continuing.
Product decision: every taught occurrence needs a real source passage and a meaning that matches it. Audio supports pronunciation, while optional illustrations provide an additional clue only when available.
Stahl & Fairbanks (1986), a meta-analysis of vocabulary instruction
Ask for retrieval, then leave room before asking again.
Trying to recall an answer changes memory differently from simply rereading it. Leaving time between those attempts generally supports longer retention better than clustering the same amount of practice together.
That is why lessons bring due words back alongside a manageable number of new ones. A successful answer allows a longer interval. A difficult answer returns sooner and can receive more support. “Spaced repetition” is the technical label; the practical idea is simpler: review when remembering takes useful effort.
Product decision: schedule each word from that reader’s history instead of assigning the same review calendar to everyone.
Cepeda et al. (2006), distributed-practice meta-analysis · Rowland (2014), retrieval-practice meta-analysis
Make the answer more exact as knowledge grows.
Knowing a word is not a single yes-or-no state. A reader may recognize its general meaning before being able to distinguish a precise sense, choose it in context, spell it, or recall it without a visible answer.
The app reflects that progression. Early meaning questions can use clearly different alternatives. Later alternatives become semantically closer. Reviews can move from recognition to contextual choice, spelling, and a letter Jumble. The aim is deeper knowledge without making the first encounter needlessly punishing.
Product decision: vary both the exercise format and the closeness of wrong answers, while aiming for a learning sweet spot of roughly 80–90% correct.
Teach word parts that can transfer.
Prefixes, roots, and suffixes connect individual words into families. Instruction in morphology has shown benefits across vocabulary and literacy outcomes, with particularly useful effects for readers who need more support.
The app introduces word parts in dedicated exercises and connects them to vocabulary already appearing in the reader’s books. This does not replace learning a word in context; it adds a second structure for understanding and remembering it.
Product decision: surface productive word parts when they add genuine explanatory value, rather than forcing every word into a decomposition.
Bowers, Kirby & Deacon (2010), systematic review of morphological instruction
Use evidence to choose priorities—not decorate a finished idea.
Our research collection is part of product planning. A promising idea earns priority when the evidence is reliable, the effect is educationally meaningful, the participants and task resemble our use case, and the mechanism can be implemented without making sessions confusing or exhausting.
- Strength: Prefer syntheses, replications, and converging results over a single striking paper.
- Fit: Check whether the ages, learning task, and outcome actually resemble vocabulary learning from books.
- Translation: Turn the mechanism into one clear product decision that can be observed and tested.
- Restraint: Do not add a feature merely because it can be described with research language.
What we are—and are not—claiming
Evidence Based Vocabulary has not been evaluated in a peer-reviewed efficacy trial. We do not claim that the app itself has been proven to produce a particular improvement, or that every research result transfers unchanged into this setting.
We do claim that the design is intentionally informed by established research mechanisms, that the implementation is checked against what the product actually does, and that we will keep revising both when better evidence or better product data arrives.