PBE Quiz Generator (Static Site)

Static HTML/CSS/JS site for generating:

• Printable quizzes
• PowerPoint quiz decks
• Interactive web quizzes

The app reads question files from questions/v1 and runs without a backend, so it works on GitHub Pages and local static hosting.

Current Features

• Filter by question type
• Filter by one or more difficulty metadata values
• Select individual verses within selected chapters
• Human-reviewed-only filter (validatedBy = human)
• Total question count target
• Number-per-verse cap
• Realtime generation as selections change
• Year selector (config file based)
• Person profiles saved in browser localStorage
• Multi-person selection for printable output sections
• Print view with question sheets and answer key sheets
• PowerPoint deck export (client-side via vendored PptxGenJS)
• Interactive quiz mode with scoring
• Fill in the Blank questions generated from the NKJV verse text (see below)
• Reorderable people list: drag, or use the arrows, to change the order
• Intro/commentary passages for books that have them
• One-click import of presets from the old PBE Study Tool
• Your own questions: download a spreadsheet, fill in a question, answer and points (book, chapter and verse are optional), and add it back. They are kept on the device (IndexedDB), can be mixed in or used on their own ("Questions from" in simple and advanced setup), and can be shared with everyone through a pre-filled GitHub issue. Questions without a reference can't be picked by chapter and show no verse text.

Run Locally

Use a local static server for normal hosted-like behavior.

Option A: Python

python3 -m http.server 8000

Then open http://localhost:8000.

Option B: VS Code Live Server

Open index.html with Live Server.

Local File Mode (No Server)

This project now supports opening index.html directly from file:// by using a generated JS data bundle.

1. Generate or refresh the bundle:

node scripts/build-local-bundle.mjs

2. Open index.html directly in your browser.

The app will use questions/v1/local-data.bundle.js automatically when present, and use fetch-based JSON loading when hosted or served.

GitHub Pages

1. Push this repository to GitHub.
2. In repository settings, enable GitHub Pages from the root (or main branch root).
3. Open the published URL.

No build step is required.

Data Files

• Manifest: questions/v1/manifest.json
• Chapters: questions/v1/by-chapter/<book-id>/<chapter>.json
• Year mapping: questions/v1/years.json
• Practice books (with intro commentary) and seasons: questions/v1/practice-books.json, practice-chapters-by-year.json

If years.json has no scope for a year, that year defaults to all books and chapters in the manifest.

years.json format

{
  "schema": "pbe.year.scope.v1",
  "version": 1,
  "years": [
    {
      "id": "2026-current",
      "name": "2026-2027",
      "scope": {
        "41-mark": [1, 2, 3],
        "60-1-peter": [1, 2, 3, 4, 5]
      }
    }
  ]
}

scope is optional. If omitted, all manifest chapters are included for that year.

Verse Selection UX

• Chapter checkboxes still control the primary scope.
• Each selected chapter has a verse picker with All verses as the default.
• Quick actions inside each chapter picker: All, Clear, and Range (start + end + Apply).
• Choosing custom verses limits generation to questions whose references overlap those verses.
• Custom verse choices are saved per person profile in localStorage.

Fill in the Blank

Fill in the Blank questions are generated from the verses themselves, so every verse in your scope can become a question. The blanked words are the most meaningful ones (names, nouns, verbs) chosen with the compromise NLP library plus TF-IDF scoring, ported from the PBE Study Tool.

• Match difficulty (default): easy blanks up to 20% of a verse's words, medium over 20-40%, hard over 40-80%. Each question picks a random percent inside its band, so verses blank differently each time. The difficulty filter and difficulty percentages work as they do for written questions.
• Set it myself: override the bands with a number of blanks or a percent of the words. Questions are still labelled easy/medium/hard by the share blanked, and only the difficulties you ticked are used.
• Points equal the number of blanks, so the PowerPoint and interactive timers scale with them.
• Written and generated questions can be mixed. With no percentages entered, a type's share follows how many questions it has (generated questions count once per verse); enter percents beside the types to set the mix.
• Generated questions skip the Question Source filter, since no author is involved.
• Intro/commentary passages appear as an Intro entry per book (fill in the blank only).
• Blanks are chosen once per quiz with a seeded random generator, so the preview, print view, PowerPoint and interactive quiz all show the same blanks.

Verse text is downloaded from bolls.life when a chapter is first needed and cached in IndexedDB, so repeat visits are instant and work offline for chapters you've already used.

Interactive quiz

Each blank is a text box in the verse; Tab or Enter moves between them and Enter on the last one submits. Tap a blank to type in it, then tap its ? once for the first letter and length and again for the whole word. A blank given away by a hint counts as incorrect in the default score (you can still adjust the points on the answer page), and the answer page notes how many hints were used. Using a hint on every blank submits the question automatically. Spelling slips are forgiven: answers are compared with Double Metaphone (sounds alike) and Jaro-Winkler (close spelling), vendored in assets/vendor/spelling.js (rebuild with npm run build:vendor-spelling; licenses in licenses/spelling/). Words that sound alike but are a different answer (Jude/Judah) or appear elsewhere in the question, verse or quiz are not forgiven. Right and wrong answers differ by weight, underline and strike-through as well as color, so they read without telling red from green. Typed answers earn partial credit by their key words (names, nouns, verbs, numbers; small words like "why are you so" earn nothing), judged in assets/js/answer-grader.js with the NLP library's word types, a stemmer (make/making) and the spelling check above. A multi-point answer is split into one part per point where it can be ("...fearful? ...faith?"), and a part needs about two thirds of its key words. The points are only a starting point: the answer card still lets you change them. In the interactive quiz the answer timer waits for the question to be read, as in PBE: it starts when the estimated reading time is up or as soon as you tap an answer box, whichever comes first (assets/js/reading-time.js). Reading speed was checked against the 2026 NAD testing broadcast: English is read at 1.93 words/s (off by about 1.2 s on average), and NAD's English-and-Spanish lead-in runs at 0.83 words/s. State testing (Florida and Custom) is English only, so its lead-in is half of NAD's; NAD and Nearpod use the full one. The fixture of counts and seconds is in tests/fixtures/. Finishing a quiz celebrates with confetti (not shown if your system asks for reduced motion) and offers Practice again with new blanks: the same verses with different words blanked.

People

Each person keeps their own year, scope, verses, filters and fill-in-the-blank settings. Drag a person (or use the arrow buttons) to reorder; names, order and settings save as you change them. Tick Print to include someone in the printed quiz.

Importing from the PBE Study Tool

The study tool links here with its presets in the URL (#import=...). The quiz builder asks before adding them as people set up for fill in the blank practice (same chapters, verses and blank counts).

Tests

npm install
npm test
npm run test:coverage

Timers

The PowerPoint and interactive timers offer Florida (10 s first point, 5 s each additional), NAD and a custom setting. The NAD competition runs on Nearpod, which times in 15-second steps. Measured from the 2025 and 2026 broadcasts (98 questions), 1-2 points ran 30 s, 3-5 points 45 s and 6-7 points 60 s, so NAD is 30 s plus 5 s per extra point, rounded to the nearest step (86% of the questions; the earlier 30 s plus 10 s, rounded mostly up, matched 59%). The counts behind this are in tests/fixtures/nad-nearpod-steps.json. A timer choice saved as "Nearpod" now loads as NAD. The interactive timer also waits for the question to be read (see above): Florida and Custom wait the English reading time, NAD waits twice as long (English then Spanish), and Custom lets you pick either.

Notes and Limits

• Difficulty filters use the difficulty field in each question (easy, medium, or hard) and can include any combination of those values.
• PowerPoint export uses the vendored assets/vendor/jszip.min.js, loaded when you download a deck, so CDN access is not required. (assets/vendor/pptxgen.min.js is still vendored but nothing loads it.)
• The included sample deck (2023-NAD-PBE-Practice-Test.pptx) is used as style direction, not binary template injection.
• If manifest.json references a missing chapter file, bundling skips that chapter and logs it.

Third-Party Licenses

• Notices and dependency metadata: THIRD_PARTY_NOTICES.md
• License text copies: licenses/

Next Enhancements

• Add optional profile export/import JSON for sharing across devices.

Question Data Pipeline

Everything derived from questions/v1/by-chapter/** is generated, not hand-edited:

• manifest.json: books, chapters, verse/question counts and a content hash per chapter
• packs/<season>.<hash>.json and packs/index.json: one pack per season with all of its chapters
• practice-years.bundle.js, practice-books.bundle.js: script-tag copies of the practice JSON files

Seasons are defined in practice-chapters-by-year.json (upcoming seasons through 2035-2036 are already listed). To add questions or a season, edit the source files and push to main: the Build question data GitHub Action regenerates and commits the rest. To do it locally run npm run build:data; npm run check:data fails when generated files are stale (the test suite checks this too).

Loading and Speed

Questions load one season at a time from questions/v1/packs/ and are cached in IndexedDB, so the first load stays the same size as the question bank grows. See docs/PERFORMANCE.md for the design, how to measure it (npm run bench) and the before/after numbers.

sw.js (the service worker) and the packs are generated by npm run build:data; the service worker only registers over https, or with ?sw=1 when testing.

To open index.html straight from disk, run node scripts/build-local-bundle.mjs first; the app falls back to those bundles when fetch is blocked.