Method
Nothing here asks to be believed. This page sets out how each analysis works, so that a result can be examined rather than accepted — and so that the places where a method is blind are as visible as the places where it speaks.
The shape of the thing
Moria is one file. All five analyses were computed in advance over the whole canon and packed into it, along with the texts themselves, so that nothing is fetched and nothing is logged. The build is a pipeline of eight steps, from reading the source corpora through to assembling the finished page; the whole of it is reproducible from public data.
The unit throughout is the verse — 31,102 of them, in the versification the King James Version uses. Getting everything onto that one grid is most of the work, and two of the eight steps do nothing else.
Getting the texts onto one grid
The Dutch text comes from the extended Statenvertaling export, which carries the marginal notes and a Strong’s number on every word, and which already follows the KJV versification. The simpler export does not: in some sixty psalms it follows the Masoretic count, where the superscription is verse 1, and so runs one or two verses out of step with everything else.
Rather than shifting those by hand, the check runs in two stages. The Old Testament project’s own VerseMap says which chapters are counted differently and how — that gives the candidates. Each candidate is then tested empirically, by comparing the Dutch against the English on shared proper names, numbers and verse length: signals that survive translation. Only where the Masoretic reading scores clearly better is it applied. The step now reports rather than corrects, so that if the text source is ever changed the problem shows up immediately instead of quietly.
The Septuagint bridge
The New Testament almost always quotes the Old in the Greek of the Septuagint. Hebrew and Greek cannot be compared word by word; Hebrew and the Septuagint cannot either — but the Septuagint and the New Testament can, because they are in the same language. So every Old Testament verse that has Septuagint text is compared directly against every New Testament verse.
Two signals are combined: rare three-word chains occurring in both verses, which is the signature of a quotation — not the same words but the same order — and cosine similarity over tf-idf of word forms and word pairs, which catches looser allusion.
The Septuagint has its own verse divisions. The psalms shift; Jeremiah is ordered differently. The alignment onto the ordinary numbering is derived from the text rather than tabulated: proper names are recognisable across languages, so Greek and Hebrew names are reduced to a consonantal skeleton and matched, and a monotone alignment runs per book with name agreement and verse length as its cost. Merges and splits — Septuagint Psalm 9 covering two Masoretic psalms — stay monotone and come out of that by themselves. Jeremiah, the one rearranged book, had two chapter mappings tested against each other, one computed and one the classical table from textual criticism, and the better-scoring one was kept. Against 34 known checkpoints — Psalm 110:1 as Septuagint 109:1, Jeremiah 31:31 as 38:31, and so on — the alignment is right in 33.
Why the evidence is recomputed, not stored
Each link is shown with the words that produced it, and those words are worked out again in the browser when you look at them, using exactly the same segmentation and stop-word lists as the original analysis. Storing the highlighting would have been smaller and faster, and would have risked marking something other than what actually caused the score.
There is a diagnostic in that. If the score is high and the highlighting comes back empty, the link is running along a different track — through the Greek, say, while the translations diverge. That is information, and it is only visible because the evidence is derived rather than remembered.
Where the methods go quiet
Every one of the five has a blind spot, and knowing them is most of what it takes to read the results honestly.
The cross-references show no textual evidence at all, because they are not derived from the text. They record where readers, over centuries, judged two passages to belong together.
The original-language method cannot cross between Hebrew and Greek. It will never connect a Hebrew psalm to its Greek quotation directly.
The Septuagint bridge finds only quotations that follow the Septuagint’s wording. Where the New Testament follows a different Greek text, or translates the Hebrew itself, the bridge is silent — and its silence is not evidence of absence.
The quotation method through the translation runs through a translation, which means the translators’ vocabulary partly determines what is found. The quotation and thematic analyses are computed on the Dutch and English texts; changing the reading translation in the interface changes what is displayed, not what was scored.
The thematic method is statistics. It knows that two texts use language the same way. It does not know what either of them means.