How an idea joins the library — Grammar of Meaning

Before the search

The library was never empty.

Before I went looking for Tengri, I'd already scaffolded it.

A hypothesis-list of traditions worth including, plus earlier passes of background research and first-draft notes, gave the library a shape: shelves with labels, some already full. So a new idea doesn't drop into a void — it lands in a structure that already exists.

→The engine adds to a partly-built library, not a blank page.

1

First, the engine checks: is it already here?

Every idea is matched against what the library already holds.

Try it — illustrative

Tengri Four Noble Truths Ubuntu Apophatic Prayer

→A miss isn't a failure. It's the case the engine exists for.

2

Not here. So the engine searches everywhere at once.

Tengri is searched across many archives in parallel — its home-region collections and the big Western ones, with equal priority.

✦ Tengri →

Mongolian manuscript archive · home region3 found

Eurasian oral-history corpus2 found

Siberian ethnographic collectionfound

Open scholarly repository · globalpapers found

Major Western theology indexempty — logged

Searching…

Illustrative of how the search fans out; counts shown are representative.

→An archive is searched for what it contains, never for whose region it "belongs to." Skipping the obvious ones is how blind spots form.

For the curious — the tools doing this

The search runs on a small stack I built:
• CRADLE — the acquisition pipeline: classify a request → route it → resolve it to real sources → dedupe → ingest.
• PRISM — tags every archive on 13 routing axes (region, language, era, peer-review…), so the engine knows where to look.
• FOUNDRY — builds the individual harvesters that pull from each archive.
• The acquisition queue — the engine's running want-list of what to acquire next (internally, the "shopping cart").
Two kinds of harvester run side by side: core (general, cross-tradition archives) and tradition-specific (e.g. steppe / Mongolian sources). 4,745 archives catalogued so far.

3

Each source arrives with a full trail.

Nothing comes in anonymously. Every source carries where it came from, how it was fetched, and who wrote it — and the empty searches are recorded too.

For a passage about Tengri, the trail records three things that can never be left blank — where it came from, what shaped it, and what wrote the row — plus the live source URL, the citation locator, and its access/consent status. So any later claim can be traced back to the exact text. When an archive comes back empty, that absence is logged, not silently dropped: over time the logged gaps become a visible map of where the world's record is thin.

✓ found · Mongolian manuscript archive ✓ found · Eurasian oral-history corpus ✗ empty · Western theology index — logged

→Every later claim traces back to a source — and the misses are part of the record.

For the curious — the fields on every row

Three provenance fields are enforced at the database layer and can never be NULL:
• where it came from (extraction_provenance) — which archive, harvested how.
• what shaped it (methodology_notes) — the scholarly / methodological context.
• what wrote the row (description_author) — the exact script or process.
Alongside them: the live source URL, the citation locator (so a claim points at the precise passage), and the access / consent status. No provenance, no entry — every coded value can be audited back to its origin.

4

Big texts become small, quotable passages.

A whole book is too big to compare. So the engine splits each text into passages — the smallest unit you could quote and stand behind. This is the step we call chunking.

"By the will of Möngke Tengri — the Eternal Sky — under whose power all things live, this decree is given."

↑ one passage, lifted from a 13th-century steppe edict. It can now be pointed at, quoted, and labelled on its own.

→The passage, not the book, is the unit we analyze. 392,000 sources have become 9.6 million passages.

For the curious — how a passage is sized

Texts are split into a shallow hierarchy:
• L0 — a shared canonical root (e.g. a scripture many traditions read).
• L1–L3 — organizational scaffolding (e.g. book → chapter → section for scripture; sermon → section → turn for a transcript).
• L4 — the quotable passage itself. "Atom" = an L4 passage — the actual unit of analysis.
Container vs. atom: the book is the container; the L4 passage is the atom. A small shelf of books yields a vast number of atoms.

5

Before coding: triage, then the tradition's own glossary.

Coding doesn't start cold. First each passage is triaged, then the engine seeds a glossary in the tradition's own words — so the analysis lands on the right material, in the right vocabulary.

Triage tags each passage cheaply first — how central is it (a 13th-century imperial edict invoking Tengri is high-significance), and what kind of artifact is it (a decree, a chronicle, a modern commentary)? — so the expensive interpretive step lands on the right material, and a commentary is never mistaken for the thing it comments on.

Glossary seed. Then, before any coding, the engine collects the tradition's own in-vivo terms — the words a member would actually use:

Möngke TengriKöke Tengri (blue sky)sülde (spirit)jayaɣa (mandate)

→Triage before expense; the tradition's vocabulary before any interpretation. The anti-pre-anchoring discipline, built into the order.

6

Each passage is coded in its own tradition's words.

No outside category is forced on. Pass 1 codes the passage in its own vocabulary, read three ways — insider, outsider, practitioner — so no single viewpoint decides. Pass 2 checks each code back against the tradition's glossary.

Insider

"Reverence for the sky as the source of legitimate rule."

Möngke-Tengrisky-mandate

Outsider

"A high-god sanctioning earthly authority — recognizable cross-culturally."

divine-sanction

Practitioner

"You invoke Tengri before acting; the sky is witness, not idol."

invocationwitness-sky

The instrument records more than a label: the in-vivo terms, a scale of concern (how wide a circle the passage cares about — from a single self to the whole cosmos), and a short orientation note (what is this passage for?).

Notice: nobody wrote "monotheism" or "God." Those are outside categories — applied later, carefully, only after the passage's own words have been heard.

→Each passage is described in its own tradition's words first — no imported categories.

For the curious — the two-pass coding instrument

Pass 1 — each of the three readers gets a prompt of this shape: Read this passage. Describe what it says using ONLY the tradition's own vocabulary — the in-vivo terms a member would use. Do NOT import outside categories yet. Role: {insider | outsider | practitioner} Return: in-vivo term codes + the scale of concern (how wide a circle it values) + a 2-3 sentence orientation note (what is this for?). The real instrument asks for more than a one-line reading — the scale of concern, the tradition's own glossary terms, and that orientation note are all part of the record.

Pass 2 — the glossary drift-check. Each code from Pass 1 is compared back against the tradition's seed glossary; terms the glossary lacks flow back into it (grounded-theory style), so the vocabulary keeps growing. The glossary is a drift-check, never a cage.

The three independent readings get per-axis agreement scoring; disagreement flags a passage as uncertain rather than being smoothed over. (This stands in for a human rating team a solo researcher can't afford.)

✦

Tengri is in the library — and it feeds the next search.

The engine isn't a one-way pipe. What it brings in becomes fuel for what it looks for next.

Coding Tengri surfaces new in-vivo terms and citations; those refill the acquisition queue; which drives the next search. The library grows itself — it's a loop, not a line.

→Bringing one idea in makes the engine better at finding the next. That's why it's an engine, not a pipe.

Where Tengri ends up: in the library — sources gathered with a full trail, split into passages, coded in its own vocabulary. Ready to be compared with everything else.

What the library then does with all those coded passages — where traditions converge, and where they don't — is a different question, and a different story. This page is just how an idea gets in.

What happens when Tengri wants in?