Project Tavnit
What shape does Scripture make when you let it organize itself? We built a network of 1.4 million biblical connections across 17 independent layers and asked the mathematics to show us the structure. It did.
Scroll
The project name comes from the Hebrew word tavnit (תבנית) — the “pattern” or “blueprint” that God showed Moses for the Tabernacle (Exodus 25:9).
We gathered over 1.4 million verse-to-verse connections across 17 independent layers — from purely computational word-frequency analysis to connections identified by over 2,000 biblical teachers and commentators spanning two millennia. We used the text's own ancient paragraph markers (the petuchah and setumah divisions transmitted by Masoretic scribes and attested in Dead Sea Scroll fragments) as our units of analysis rather than modern chapter-and-verse divisions.
Then we applied network topology algorithms — mathematical tools borrowed from genetics, social science, and physics — to let the connections reveal the text's own structure without imposing any framework from outside.
When the data contradicted our initial hypothesis, we reported the rejection. When a control test caught an overreach, we corrected it. We’ve dedicated a section below to what we got wrong — because the methodology’s capacity to say “no” is what makes the surviving “yes” answers credible.
Every non-Torah book in the Bible — all 61 of them — has its strongest vocabulary connections pointing to the same region: the Sinai legislation (Exodus 18 through Leviticus 27). True in Hebrew and Greek, confirmed by three independent computational methods, with bootstrap confidence intervals so tight that chance is not a plausible explanation.
The text's connection network organizes along 13 independent interpretable axes. Each corresponds to a recognizable theological tension: covenant vs. worship, warning vs. execution, sacred space vs. sacred time. These fall into four categories: substrate axes recovered by multiple independent methods, tradition-specific axes that depend on which interpretive tradition built the bridge, tradition-engaged axes where commentary traditions actively participate, and a genre bedrock axis that no commentary tradition touches.
Five medieval Jewish commentators tested individually on all 13 axes. Rabbeinu Bahya (14th c., kabbalistic method) perturbs PC10 with extreme force. Peshat commentators — Rashi, Ibn Ezra, Abarbanel — reinforce the same axis. Profiles cluster by exegetical method, not by century. The “tradition divider” axis is kabbalistic-vs-everyone-else, not Jewish-vs-Christian.
Participation ratio 18.1 (Greek literature: 6.0). The same pattern holds when tested in the Septuagint — the signal transcends language. Persistent homology reveals 129 structural loops versus 6 in comparable Greek corpora (21×). The Bible’s connection network follows a power-law scaling (N2.7) that Greek literature lacks entirely at any scale. This is not a property of ancient texts in general. It is specific to this corpus.
The algorithm recreated Torah, Prophets, and Writings from connection topology alone. Nobody told it these categories existed.
The Tokhekhah is the most multi-dimensionally significant passage in the Hebrew Bible. The text organizes around covenant consequences more than promises.
Feast passages encode direction, not mass. Spring feasts connect to narrative fulfillment; fall feasts to interpretive theology. Spin, not gravity.
The New Testament engages only the covenant axes. Worship, sacred space, and history axes are not engaged at statistically significant levels. The original 11/11 result was an overreach — controls corrected it.
Two axes are reinforced by every interpretive tradition tested — Jewish, Christian, and computational. PC8 is universal among medieval Jewish commentators regardless of method.
The worship axis converges across three independent layer families: co-citation (TSK), scholarly allusion (Beale-Carson), and raw vocabulary (Greek). The bridge is priestly vocabulary, not messianic prophecy.
The Sinai attractor independently rediscovered by Meredith Kline’s covenant theology (90.7% structural overlap) and Alexander Poltorak’s Torah physics (99.1st percentile in our null distribution).
Jackknife resampling across 26 teachers. Convergence rate and cross-tradition agreement are tight enough to rule out chance but wide enough to acknowledge real uncertainty.
Before Nicaea, Church Fathers perturb 3 axes and reinforce none. After Nicaea, they reinforce 6 and perturb none. A clean era-specific divergence in how the Christian tradition structurally engages the OT.
Every finding above survived a methodology designed to kill bad ideas. These are the ideas it killed.
We went looking for a double-helix structure in the Torah’s topology. The data rejected it with a correlation of 0.009 — functionally zero. We found a tree instead. The helix was our hypothesis; the tree is what the mathematics showed.
Our initial analysis suggested the New Testament engaged all 11 structural axes. Permutation controls revealed that 8 were majority-dominance artifacts — the NT contributes so many connections that it distorts weak axes. Three genuine axes survived: all in the covenant grouping.
A searchsorted truncation error in our broken-stick implementation
reported 11 significant components when the correct answer was 1–2 dominant axes.
All 13 validated axes are real — the interpretations were built from actual
loadings, not the buggy count — but the headline metric was wrong.
We caught it, logged it, and fixed it.
Matthew Henry’s commentary, tested by both co-citation and LLM extraction, produced opposite results on PC9/PC10. Investigation revealed that co-citation perturbation is density-dependent — an artifact of method, not meaning. This led to density-matched controls that now protect every per-author null test. The correction made the methodology stronger.
The project's credibility rests as much on what it falsified as on what it found.
Every non-Torah book's connection center of gravity falls in the same region — the Sinai covenant legislation of Exodus through Leviticus. 61 books, three independent methods, two languages, one basin.
2D / 3D · Three.js115-node parashah-level network with layer isolation, weight filtering, and feast highlighting. Rotate, zoom, and explore the connections between Torah portions and biblical books.
3D · Three.js3,151 ancient paragraph nodes in PCA space. Browse the text's own topological structure without any imposed coordinate system.
3D · Three.jsConceptual atlas of the 13 PCA axes — four categories of structure (substrate, tradition-specific, tradition-engaged, genre bedrock), which passages define each pole, and how traditions engage each axis. Being updated to include PC14 and PC19.
3D · Three.jsProject briefing with connections to existing scholarship (Fishbane, Levenson, Heschel, Hays) and open questions for theological engagement.