Recognizing a word is the easiest part of knowing it — most vocabulary apps stop there. LexForge makes you produce it.
On the first screen, click Try with sample words to go from recognition to writing your own sentence in one click. You judge your own sentences: free production is self-checked, and nothing in the app grades it.
Recognizing a word is only one part of knowing it. LexForge tracks whether you can retrieve it, distinguish it, use it correctly, and produce it in writing.
A word isn't Exam Ready just because you recognized it — LexForge requires successful independent production and novel application.
- ✍️ Mastery requires real production
- 🧠 11 dimensions of vocabulary knowledge
- 🔎 7 categories of mistake diagnosis
- 📈 FSRS-6 scheduling for recognition, meaning recall, and production
- 🔒 Fully local — no account, no server, no network
| Conventional flashcards | LexForge |
|---|---|
| Recognition can feel like mastery | Mastery requires production |
| One-dimensional card state | 11 skill dimensions |
| Wrong answer → mark incorrect | Diagnose the type of mistake |
| Review at the word/card level | FSRS-6 for recognition, meaning recall, and production |
| Practice centers on retrieval | Teach → retrieve → produce |
| In-app exercises | Writing Audit tests real usage |
Anki is a mature, general-purpose spaced-repetition tool, and recent versions can schedule with FSRS too. LexForge doesn't claim to out-schedule it, or that recall practice is new: an Anki reverse card (meaning → word) is already cued recall, on its own schedule.
What LexForge adds is a stricter definition of "known" for vocabulary:
- Production unlock gating. A word's FSRS production card only unlocks once its recognition stability passes a threshold. (The older level-based ladder, which also reaches free production, runs alongside it.)
- Free-production prompts. You get a situation and a constraint and write your own sentence, instead of flipping a card.
- Mistake diagnosis. A miss is sorted into one of seven categories (Meaning, Form, Spelling, Grammar, Collocation, Register, Selection) and tied to the skill it affects.
- Exam Ready needs production. A word reaches Exam Ready only after independent production and novel application have succeeded, not on recognition or recall alone.
Free production is self-checked. The app shows your sentence next to the word's meaning, collocations, grammar, and contrast, and you mark whether it was right. There is no AI grading and no automatic check, so the result is only as honest as your own judgment.
Where Anki is the better choice: a huge ecosystem of decks and add-ons, mobile apps, and sync. LexForge is one browser's localStorage (export and import by hand), and it only does vocabulary.
I kept noticing a problem with vocabulary apps: recognizing a word feels like knowing it, but recognition and actual use are very different skills.
A student can recognize substantiate in a multiple-choice question and still fail to retrieve it, distinguish it from prove, use the wrong preposition, or avoid using it in writing entirely.
So I built LexForge around the idea that vocabulary mastery should end in production, not recognition.
Teach → Retrieve → Produce → Diagnose → Schedule → Apply in writing
- Teach. Every new word goes through a guided eight-step flow, from introduction and contextual examples to constrained and independent production. No word enters practice until it has been taught.
- Retrieve. Recognition, cued recall, and cloze questions check that you can bring the word to mind, in a Daily Session, a due-only Quick Review, or targeted drills.
- Produce. Free production gives you a situation and asks you to write your own sentence with the word, then check it yourself.
- Diagnose. A mistake isn't just marked wrong. It's sorted into one of seven categories (Meaning, Form, Spelling, Grammar, Collocation, Register, Selection) and tied to the skill dimension it affects.
- Schedule. Recognition, meaning recall, and production each get their own FSRS-6 card per word, and production cards unlock once recognition is stable. The other dimensions use an older level-based ladder.
- Apply in writing. The Writing Audit lets you paste your own writing and tag errors against words in your bank, and matched errors feed back into scheduling. A word reaches Exam Ready only once independent production and novel application have both succeeded.
The workflow, from the dashboard to a production question to analytics.
The screenshots use a synthetic data set generated for this README, not real study data.
For a full walkthrough with screenshots of every screen, see docs/WALKTHROUGH.md.
Teaching phase
- Guided eight-step flow for each new word: introduction, core meaning, contextual examples, guided processing, supported retrieval, constrained production, independent production, entering retrieval.
- Add words one at a time, or paste many at once in a pipe-delimited format. The app includes a prompt you can give an AI assistant to generate that format.
Practice engine
- Daily Session, Quick Review (due items only), and context/production drills.
- Question types cover recognition, cued recall, cloze, and free production.
- Each word is tracked across eleven dimensions: form recognition, form recall, meaning recognition, meaning recall, semantic discrimination, grammar, collocation, contextual comprehension, guided production, independent production, and novel application.
- Mastery stages are New, Learning, Unstable, Stable, and Exam Ready. Exam Ready requires successful independent production and novel application, not recognition alone.
Scheduling (FSRS-6 + legacy ladder)
- Recognition, meaning recall, and production are scheduled by FSRS-6 (via the vendored
ts-fsrs), one card per word and skill. - Production cards unlock once recognition stability passes a threshold.
- The other dimensions still use the original level-based ladder with fixed intervals. The two systems run side by side.
- An optional exam date compresses review intervals for words that are not yet Exam Ready.
Writing Audit + Timed Exam
- Weekly Writing Audit: paste your own writing and tag errors against words in your bank. Matched errors feed back into scheduling.
- Timed Exam: write to a prompt under a timer with target words, then mark which ones you used correctly.
Analytics
- Weekly activity versus the prior week, streaks, achievements, and a milestone list.
- Words added versus words reaching Stable, mastery distribution, upcoming review load, and a per-dimension coverage table.
- Scheduler diagnostics: prediction calibration, stability growth, retention by interval, and an activity heatmap.
Privacy
- All data lives in your browser's
localStorage. There is no server and no account. - Export and import (merge or replace) as a JSON file from Settings.
- Everything the app needs, including Chart.js and the FSRS library, is vendored in
lib/. The app makes no network requests.
Live demo: https://da-boss-is-here.github.io/Lexforge/
First run: the bank starts empty. Try with sample words loads five demo words (flagged as sample data, removable from the Dashboard) and starts a short recognition → production round without the Teaching flow. Or add your own words.
Run locally: open index.html in a browser. No server, no build, no install. The app is one HTML file plus the libraries vendored in lib/.
node --test tests/*.test.js
Requires Node.js. The glob is needed; node --test tests/ alone finds nothing.
index.html— the whole app: markup, styles, and one inline script.lib/— FSRS scheduling and due-queue modules, shared by the app and the tests, plus the vendoredts-fsrsand Chart.js builds.tests/— unit and smoke tests. The persistence smoke test loads the real script out ofindex.html.assets/— the README demo GIF/MP4, regenerated byscripts/capture-demo.js(dev-only; needs Playwright and ffmpeg, which the repo doesn't depend on).docs/— the walkthrough and the screenshots used by it and the README.
See CONTRIBUTING.md to work on it, and CONTEXT.md for a file and function map.
MIT. See LICENSE.
The root MIT license covers the original code in this repository. The vendored ts-fsrs and Chart.js builds in lib/ carry their own licenses — see lib/fsrs.LICENSE, lib/chart.LICENSE, and THIRD_PARTY_NOTICES.md for details.
Open bugs found during code review are tracked in KNOWN_ISSUES.md.