Every Promise

A Bible built to be two things at once, where both are required:

Easy to read — can you follow it without stopping? Honest underneath — does every English word describe the Greek word beneath it?

A beautiful sentence with the wrong word under it is a broken product, not a stylistic quibble. Everything in this repository exists to keep both true at the same time.

Everything here is open source — the app, the four texts, the lexicons, the hymnal, the tools that assembled every file, and the commit history recording each correction and the reason for it.

Live at https://everypromisebible.com · on Google Play as Every Promise.

---

The four texts

The Mak Translation — tx.01

A word-for-word English New Testament laid over the SBLGNT. Every English word sits on the specific Greek word it came from, so tapping a word tells you what that word means rather than what the sentence roughly means.

Each word renders as a three-line stack — the English, its Greek, and the Strong's number with the grammar. Words supplied for the sake of English, with no Greek behind them, are set in black italic, so a reader can always see what came from the text and what did not.

The English is arranged in English order rather than Greek order, which is what makes it readable. The Greek order stays recoverable from the tags, and the invariant that must never break is that every Greek word appears exactly once.

Chapters	260 — the complete New Testament
Sections	1,190
Word units	117,353
Greek tokens	137,554

The Illumination Translation — tx.02

A different job entirely. Where Mak is anchored to the Greek word by word, the Illumination is free prose: a loose, thought-for-thought paraphrase of the whole Bible in original modern English, composed from the public-domain 1769 Authorized Version.

It is built for a reader who finds older translations hard going — long passages with no signposts, sentences whose shape has drifted out of modern English. So it is organised by subject. Every chapter carries a title, every section a heading, and verses are grouped into blocks of a single thought rather than chopped into numbered fragments. Each book opens with a synopsis of its own.

Books	66 — Old and New Testament
Chapters	1,189
Sections	5,282, each with its own heading
Verse blocks	9,031
Book synopses	66

On its underlying text. Because it was composed from the 1769 Authorized Version, it follows the Greek text the King James translators used. The Mak interlinear follows the SBLGNT, a modern critical edition. The two differ in a handful of well-known places, so a verse can read differently here than in the interlinear. That is a difference of underlying manuscripts, not of carelessness, and it is stated in the app as well as here.

The King James Version — tx.03

The Authorized Version of 1769, unedited: 66 books, 1,189 chapters. It is here so the older wording is always a tap away, and so the Illumination can be read against the text it was made from.

The Thirteenth Disciple — dv.01

Not a translation but a devotional — a gospel narrative in thirty-one days, walking beside Jesus from the Jordan to the Ascension. Original work, roughly 32,000 words.

Each day carries the harmony of gospel passages it draws on, so a reader can go and read the accounts being retold rather than taking the retelling's word for it. Days are narrative prose, read straight through rather than looked up.

---

Around the texts

Promises — 2,226 verses, each with 15 short meditations, hand-tagged across 17 moods so they can be filtered by how a reader is actually feeling. Every verse is checked against the KJV rather than trusted: tools/kjv.py --verify.

Hymns — 1,323 from The Christian Hymn Book (Cincinnati, 1870), public domain, 25,154 lines. Indexed by first line, as hymnals have always been.

Journal — a month calendar, six reading plans computed from the corpus so they cannot drift out of step with it, a daily reflection, and sermon notes with formatting.

My Stuff — highlights, notes, bookmarks, journal entries, favourites and the meditations a reader wrote, each on its own page, each linking back to the exact word or verse. And a backup file: everything written, in one .json the reader keeps.

Nothing a reader writes leaves their device. No accounts, no sign-in, no analytics, no tracking.

---

Why the Greek can be trusted

Rearranging English is exactly what can leave a label sitting on the wrong Greek word. So the claims here are checkable rather than asserted.

	Greek here	Greek in SBLGNT
All 27 books	137,554	137,554

Every book matches individually. Not one Greek word has been added, removed, or moved between chapters.

python tools/wordcounts.py     # Greek word counts vs SBLGNT, book by book
python tools/kjv.py --verify   # every promise verse against the KJV
python tools/swap_scan.py      # finds English labels on the wrong Greek word
python tools/prose_scan.py     # mechanical prose defects
python tools/hymns_build.py    # rebuilds the hymnal from its source

CHANGELOG.md records every correction, including the mistakes found in this project's own earlier work — a whole sentence of Paul's that had been overwritten, four pairs of glosses wearing each other's labels, a name tagged as the wrong person, three misquoted verses. They are documented rather than quietly fixed, because a translation you cannot audit is one you have to take on faith twice.

What is in here

index.html              the entire app — no framework, no build step
data/chapters.js        the Mak interlinear (260 chapters)
data/illumination.js    the Illumination (66 books)
data/kjv.js             the Authorized Version, 1769
data/devotional.js      The Thirteenth Disciple (31 days)
data/hymns.js           1,323 public-domain hymns
data/promises.js        2,226 promises, 33,390 meditations
data/lexicon.js         Strong's / Thayer's — 5,367 entries
data/thayer.js          the full Thayer's entries
data/abbott.js          Abbott-Smith — 5,340 entries
data/wordpictures.js    word pictures
data/commentary.js      commentary notes
data/wordcounts.js      generated proof table
tools/                  the checking and build tooling (Python)

About 27 MB of data behind a 230 KB app. The service worker caches it, so after one visit the whole thing works with no connection.

Running it locally

python -m http.server 8000

Then open http://localhost:8000. Do not just double-click index.html — the data loads from a subdirectory and some browsers block that on file://.

Deployment is in DEPLOY.md. The Android app is a Trusted Web Activity around this site, so publishing here publishes there: no rebuild, no store review.

How it was made

The Mak Translation, the Illumination and The Thirteenth Disciple were produced with the latest LLMs, working against the SBLGNT, MorphGNT, Strong's, Thayer's and Abbott-Smith. The work spans many sessions and successive model versions. The tooling in tools/ exists because a language model will produce fluent, confident, wrong output, and fluency is not evidence. Every check in this repository was calibrated against known defects before its verdicts were trusted.

As these models improve, the text can be revised again — the checks that protect the Greek are automated and re-runnable.

Honest limitations

• The prose has had one full pass. All 1,190 Mak sections were read and repaired, but a second reading would find more.
• The invisible defect class is narrowed, not closed. swap_scan.py detects English labels that have swapped places, but cannot see a swap where the wrong label is a common word, and it only looks within a small window.
• The Illumination has had no equivalent audit. It has no word-level Greek anchoring, so the checks that protect Mak do not apply to it and cannot.
• Meditation quality is guarded by reading, not tooling. The validator checks structure — 15 entries, no duplicates — not whether a line is good.
• Some editorial choices are deliberate and arguable. A few verses quote a contiguous clause rather than a whole verse; five KJV quotations keep modernised archaisms. CREDITS.md states these.

Corrections are welcome. If something reads wrongly, or a label sits on the wrong word, open an issue.

License

Part	License
Code (tools/, the app)	MIT — LICENSE
Translation and data (data/*.js)	CC BY-SA 4.0 — LICENSE-DATA

They differ because the morphology derives from MorphGNT, which is ShareAlike — so this text cannot be enclosed. Anyone may use, improve or build on it, but no one can take it, change it, and lock the result away.

Full attribution and the statement of modifications: CREDITS.md.