§19.1 Why the addressing model follows from the method
Everything in Parts I–IV has an architectural consequence, and the derivation is worth stating because it explains why the platform is built the way it is rather than in the ordinary way.
If a passage carries several simultaneous readings (§5), and each reading is the work of a named human being who can be right or wrong (§0.4), and the strength of a reading depends on which commentator attests it (§17.5), then a data model that merges root text with commentary destroys exactly the information the method depends on. A “text with notes” is not a lossy representation of the corpus; it is a representation that has deleted the thing the method operates on.
Hence three architectural commitments, each traceable to a section above:
- Immutable (from §3.2, chain of custody) — an address, once assigned, is never reissued. A citation made today resolves to the same node in four hundred years. Reassignment would be a silent corruption of the record, undetectable by the reader.
- Machine-readable (from §0.5, locators) — resolvable by any citation system, so scholarly apparatus outside the platform can point into it. That is the only way a citation infrastructure becomes infrastructure rather than a product.
- Non-conflating (from §5, §17.5, M17) — mūla and each named commentary separately addressed. A reader can cite Madhva's reading of a verse without citing the verse, and vice versa, and a machine can tell the difference.
§19.2 The layer scheme (L01–L07)
Every textual node in the Ārk Engine is addressed across seven distinct layers:
- L01 · Mūla (Devanāgarī) — the root text. No modern author; public domain.
- L02 · IAST transliteration — mechanical, verified by a named editor.
- L03 · Anvaya (word-by-word) — named editor. Records parse ambiguity where it exists (§17.1, step 4).
- L04 · Literal translation — named editor, original to this platform. Never a translation under third-party copyright, in any layer, including in examples and fixtures.
- L05 · Classical bhāṣya — one node per commentator. Named ācārya. Never merged; never summarised into an anonymous consensus.
- L06 · Modern commentary and synthesis — named modern author, separately addressed, permission recorded (§11.4).
- L07 · Editorial apparatus — grading, variants, open-register links. Platform, dated and versioned.
§19.3 URN format specification
URN Syntax Specification
urn:arka:{work}[.{locator}][/{layer}[.{contributor}]][#{segment}]
Four separators, each doing one job and no other: . descends the work's own numbering, / selects a layer, a second . names the voice that authored that layer, and # anchors a segment inside it. Every component to the right of the work is optional, so dropping components from the right always yields a valid, less specific name — which is how the breadcrumb, the “go up” action and citation shortening are implemented, as pure string operations that never consult a database.
Example: names the mūla of Bhagavad-gītā 15.15. names Madhva's commentary on it — a different node, separately citable, never returned by the first address.