Editorial Method
How This Greek New Testament Is Built
Scriptorium presents a documented eclectic edition of the Greek New
Testament, built on published scholarly data, with an open apparatus.
Every reading carries its witnesses, its edition attestation, and the
reasoning surface on which a decision was recorded. Nothing is hidden
behind the display — and nothing is claimed beyond the evidence.
Source data
The base text stream is the SR-GNT
(© Alan Bunning / Center for New Testament Restoration, CC BY
4.0, 137,361 words). Edition reading facts come from TAGNT
(STEPBible / Tyndale House Cambridge, CC BY 4.0, 142,096 words),
tracking eight editions — NA28, NA27, Tyn, SBL, WH, Treg, TR,
Byz — strictly as facts (“edition E reads X
here”). Manuscript evidence comes from the CNTR MES
transcriptions (CC BY-SA 4.0, 188 witnesses). The lexicon is TBESG
with full LSJ definitions attached (both CC BY 4.0, Tyndale House).
NA28 and SBLGNT text is never embedded: where NA28 brackets
a passage we cite that fact; where SBLGNT is compared, only its
edition flag is used.
Evidence-first apparatus
The apparatus carries 11,644 edition-dividing variant points over
4,720 verses — 3,852 substitutions, 6,042 add/omit units, 1,750
word-order cases. A single deterministic alignment grid (Needleman–Wunsch)
decides what every screen renders: 7,011 units attach to word rows,
4,426 to witness-only ghost rows, and 207 surface as verse-level
notes — attached + notes = 11,644, zero silent
drops. What the user selects is recorded, exported, and
never silently overruled.
Witness additions, resolved by evidence tiers
Words manuscript witnesses read beyond the base text surface as
their own apparatus units (4,753 groups). A seven-tier deterministic
resolver — nomen sacra, suspension, exact form, lemma,
cross-book form, guarded itacism, curated supplement —
enriches 4,572 of them (96.2%) from the corpus’s own evidence.
The 181 blanks are honest by design: each carries a researched
verdict, and ambiguous forms ship candidate pickers that lay out
every attested probability — the bare form itself is not
selectable.
Verification gates
The whole view-model is rebuilt deterministically for all 27 books
and hashed (Phase A reconciliation, model
51d49df3…); the committed
baseline fails the build on drift. A risk-class audit (Phase B)
re-derives every count from the raw data and checks each risk class
— anchors, edition lists, partial additions, homograph
pickers, supplement entries, resolver guards. The served corpus is
pinned by a versioned SHA-256 snapshot
(scriptorium-corpus-20261030T145042Z), and the raw
sources are boot-verified against a 397-file hash ledger.
What we claim — and never claim
Claimed: a documented eclectic edition built on
published scholarly data with an open, exported apparatus; recorded,
transparent editorial decisions. Never claimed: the
autograph — like NA28 and the ECM, we hold to the concept of
the initial text (Ausgangstext); equivalence with NA28; an
exhaustive apparatus; a replacement for the published editions,
Metzger, ECM, or IGNTP; or any theological authority. The text and
the apparatus are free to read; the workbench around them is the
subscription.
Erratum policy
Findings are logged before anything is changed: the public
ERRATUM.md ledger records every defect
a deterministic audit surfaces, with its class, severity, and
corrective action. Restores are never silent (the boot canary logs
its outcomes). For the most controversial reading classes
— whole disputed text blocks, theologically loaded loci,
edition-dividing content substitutions — credentialed external
review is budgeted, on the precedent of the SBLGNT and the Tyndale
House GNT.
Data version
scriptorium-corpus-20261030T145042Z
· corpus
1215e2460145…
· view-model
51d49df30474…
Reproducible from scripts/qa13–qa18 and
data/verification/.