Bible Companion — Synoptic Diff
Open the app →
A multi-level Bible study tool. The premise: every Bible app ships the same flat
reader, so this one reads a passage at four depths and treats the gospels as
something you can diff.
The four layers
|
|
|
| L1 |
Historical |
What can actually be established — archaeology, dating, geography, what's disputed |
| L2 |
Literal |
What the text plainly says, and what it asks of a life |
| L3 |
Symbolic |
The pattern beneath — recurring images, numbers, structure |
| L4 |
Spiritual |
The path of becoming, readable without requiring belief |
This is not invented. It maps onto PaRDeS (Peshat / Remez / Derash / Sod) and
the Christian Quadriga (literal / allegorical / tropological / anagogical) —
how scripture was read for roughly 1,500 years. The remix here is splitting the
literal into a historical layer and a plain-text layer, because "did any of
this actually happen?" is the real barrier for a modern reader and no Bible app
answers it honestly.
What's built
index.html — all four gospels. This is the built app.
- 209 pericopes covering every verse of Matthew, Mark, Luke and John.
Machine-verified at 100% verse coverage, zero overlaps.
- Structural diff — what each gospel places either side of a given unit.
Reordering becomes the finding rather than diff noise.
- Textual diff — LCS word alignment across the gospels that carry a unit,
in English or Greek. Diff the Greek and you're comparing evangelists;
diff the English and you're comparing translators.
- Word study — lemma search across 64,499 Greek words, with per-gospel
distribution and clickable occurrences.
- Map — 43 places with real coordinates, elevations and certainty flags;
7 toggleable journeys; elevation profiles measured against sea level.
Every place links to the pericopes that name it.
- Material world — 28 entries across flora, fauna, money and objects, with
47 tracked lemmas and live counts. Selection rule: an entry earns its place
only when the real thing changes how the passage reads.
- Reader — type a reference (
Matthew 19, mk 8, John 3:16,
Matthew 5:1-12) and read it straight through, in English or Greek. The
reader marks where each pericope begins and prints its parallels beside
the seam, so the structure of a chapter is visible while you read it.
Anything that isn't a reference is searched as text across all 3,779 verses.
prototype.html — the original two-passage prototype (Fig Tree & Temple,
the Two Feedings). Kept because it's a tighter read of the core idea.
The material world
Darnel (Lolium temulentum) is visually indistinguishable from wheat until it
heads, which is why the parable says to leave it — you cannot identify it early,
let alone pull it. Mustard is a weed that reaches 3 m from a 1 mm seed. The
Syrophoenician woman is offered the diminutive kynarion, a house dog, not
the insult kyōn. Greek has three words for "net" — a one-man cast net, a
sweeping dragnet, and a generic one — and the kingdom parable uses the dragnet
precisely because it takes everything. English prints one word for all three.
Two things the data proved
The baskets never cross. κόφινος appears 6 times and σπυρίς 4 times across
the whole four-gospel corpus. Every kophinos belongs to the feeding of the
5,000; every spyris to the 4,000. Zero exceptions. English prints "basket"
both times, so the distinction — twelve for the tribes of Israel, seven for the
nations — is invisible in every English Bible app that exists.
23 verses have no Greek, and they're exactly the famous textual variants:
the pericope adulterae (John 7:53–8:11), the angel at Bethesda, Mark 7:16,
9:44, 9:46, 11:26, 15:28, Matthew 17:21, 18:11, 23:14, Luke 17:36, 23:17.
The app flags each as present in the Byzantine/KJV tradition, absent from
SBLGNT rather than hiding it. This fell out of comparing two datasets; it
wasn't a planned feature.
Build
python3 build.py # app_src.html + data/*.json -> index.html
Everything is inlined into a single file — the artifact CSP blocks same-origin
fetch, so external data loading isn't available. No dependencies, no bundler,
no server. Open index.html in a browser.
Regenerate the derived datasets:
cd data
python3 harmony.py # -> harmony.json (the alignment table)
python3 places.py # -> places.json (gazetteer + journeys)
Data sources and licences
| Dataset |
Source |
Licence |
| English text |
World English Bible, via bible-api.com |
Public domain |
| Greek text |
SBLGNT (Logos Bible Software & SBL) |
CC BY 4.0 |
| Greek lemmas/morphology |
MorphGNT |
CC BY-SA 3.0 |
| Coastline, lakes, rivers |
Natural Earth 10m |
Public domain |
| Pericope divisions |
Robertson (1922), Stevens & Burton (1904), Eusebian canons |
Public domain |
| Harmony, gazetteer, material-world notes |
Written for this project |
— |
Full terms, including the required attribution string, are in
NOTICE.md. Project source is MIT — see LICENSE.
⚠️ MorphGNT is share-alike. The lemma data embedded in index.html carries
CC BY-SA 3.0, which propagates to anything distributing it. If this ever ships
commercially, either honour the SA terms or swap the morphology layer for a
permissive source.
Colour
The four gospel colours are validated, not chosen by eye. The first palette
(rubric red for Mark, gold for Matthew) failed: ΔE 12.6 in normal vision and
4.2 under deuteranopia — Matthew and Mark were near-identical, in an app whose
whole purpose is telling them apart. Current set passes lightness-band, chroma,
CVD-separation, normal-vision and contrast checks in both light and dark:
|
Light |
Dark |
| Matthew |
#1263A8 |
#4A90D9 |
| Mark |
#B8500A |
#CE7430 |
| Luke |
#0A7D5A |
#2BA075 |
| John |
#9B4F96 |
#B96FB3 |
Known limits
- Four-layer commentary covers two studies. 209 pericopes × 4 layers is
~200,000 words that all have to be good; generic filler would kill the thing
that makes this work. The app says so plainly instead of faking coverage.
- The harmony is one editorial opinion. Where a pericope begins and ends is a
judgment call and I made 209 of them; Robertson and Aland disagree with each
other in places.
data/harmony.py is one line per pericope, readable and
editable.
- Journey routes are schematic. The gospels give sequences of places, not
roads. Two journeys (Infancy, Resurrection) are routes the gospels actively
disagree about, and the app says so.
- Place links find pericopes that name a place, which is not always where
the scene is set.
- Lemma counts are verses, not word instances. A lemma appearing twice in one
verse counts once (τάλαντον: 14 words across 8 verses). The UI says "verses".
Next
- Ground L1–L3 in the public-domain commentaries (Matthew Henry, Keil &
Delitzsch, Pulpit) so they're citable rather than invented. Keep L4
hand-written — that voice is the part that can't be outsourced.
- Gergesenes vs Gadarenes. Matthew names one place, Mark and Luke another,
for the same exorcism — and Gadara sits 10 km inland with no cliff, while
Kursi is on the shore with one. First case where the diff engine and the map
argue with each other.
- The symbolic web (L3), the git-graph timeline, and the relationship graph —
all queries against this same dataset rather than separate apps.