Addresses
Every verse has a name that does not move
A book’s title does not change when it moves to a different shelf. A URL is the shelf: it says where a thing is kept, by whom, on what machine, under what company, and when any of that changes, the link breaks and the citation stops meaning anything. A name is not a location, and that difference is the whole of what follows.
The problem
A citation is a promise, and the promise keeps breaking
This is not a Vedic problem or even a humanities problem. It is measured, and the measurements are bleak: of roughly 3.5 million articles surveyed in one PLoS ONE study, one in five links no longer resolved, and among the oldest cohorts seven in ten had gone. More than 70% of links in law-review articles and around half of those in United States Supreme Court opinions no longer reach what they cited.
The failure mode is always the same. The citation recorded a location, the location was somebody’s server, and servers are owned by organisations that reorganise, get acquired, run out of money, or simply tidy up. Nothing was destroyed on purpose. The address just stopped being true.
The Vedic corpus has been paying this cost far longer, and in a harder currency. A verse appears under one number in one printed edition and a different number in another, both defensible, both circulating, both cited. There is no server to blame and no link to rot: the reference was ambiguous from the moment it was made.
The correction
A name, not an address
So the identifier here does not say where anything is kept. It says what a thing is: which work, which verse, which part of that verse, whose part, and which paragraph inside it. Nothing in it refers to a machine, a domain, a directory or a company, because every one of those is a fact that can change while the verse stays exactly what it was.
It resolves on your machine, against the books you have. Nothing is fetched. There is no service to be up, no account to be current, and no organisation whose continued existence your citation depends on. A name that needs a server to mean something is an address wearing a different word.
The grammar
What a name is made of
Four separators, each doing exactly one job: a dot descends the work’s own numbering, a slash selects a part, a second dot names the voice that wrote that part, and a hash anchors a paragraph inside it.
urn:arka:npt.2.9/L06.av#s1
urn:arka:The namespace- Which registry is being asked, the way a DOI prefix says which registrant issued an identifier. Nothing outside this registry can mint one.
nptThe work- A short permanent slug, at most sixteen characters, issued once and never reissued. That is what makes a ten-year-old citation still land.
.2.9The number- Structural rather than physical, section two, verse nine, so it does not move when the book is reset, reprinted, or set in a different size. No leading zeros: 09 is not a spelling of 9, it is invalid, because one object under two spellings forks the whole citation graph.
/L06The part- Which stratum you want: the verse, the transliteration, the word-by-word gloss, the translation, the commentary. Ask for the verse and you get the verse, not a page containing the verse. The number is global, L06 means commentary in every work in the registry, permanently.
.avThe voice- Whose part, where more than one exists. Two commentaries on one verse are two names, never one merged block of prose with a credits line underneath.
#s1The paragraph- A named anchor inside that part. Anchors are authored at ingestion, never computed from position, a computed index moves the instant an editor splits a paragraph, and every annotation in the corpus then points one paragraph too far, silently.
The namespace that deliberately does not work
There is a second prefix, urn:arka.reg:, held for third-party registrants and deliberately unroutable. It exists so an identifier can be recorded before it is publishable without anybody mistaking it for a live name. A reserved name that quietly resolved to something would be worse than no reserved name at all.
Truncation
Dropping parts from the right always leaves a valid name
This is not a convenience. It is how the breadcrumb, the go-up action and citation shortening are all implemented, as pure text operations that never consult a database and never ask anything.
urn:arka:npt.2.9/L06.av#s1one movement of that commentaryurn:arka:npt.2.9/L06.avthat commentator on that verseurn:arka:npt.2.9/L06the commentary, whoever wrote iturn:arka:npt.2.9the verse, every part of iturn:arka:npt.2the sectionurn:arka:nptthe book
The converse does not hold. A name with a gap in the middle, a paragraph with no part named, is well formed but can never resolve, because a paragraph is authored on a part. It gets its own sentence rather than being lumped in with a typo.
Resolution
A name that cannot open says why
Four answers, and the reader is given them as sentences. The codes beside them are the machine’s spelling of the same four facts, shown here because this is the page where that is the point, they never appear on the reading screen.
- It opens
200The work is in your library and the name resolves to something you own.
- Not in your library
404 work_absentA real book, correctly named, that you do not have yet. Not an error and not a dead link, the reader is told which book it is, so the gap is information rather than a failure.
- Not yet released
503 rights_withheldThe part exists and is not yours to read. Rights are held per part, not per book, which is why a public-domain verse stays readable when the commentary over it is not.
- Numbered two ways
409 numbering_conflictTwo editions number this passage differently and both are defensible. The reader is shown both rather than being handed whichever one the software preferred. This is the founding problem of the whole project, and refusing to silently pick a side is the point of having addressed it at all.
The parts
What the numbers mean, and why one of them is empty
Part numbers are global and permanent. They are properties of the registry, not positions in whatever edition happens to be open.
| Number | What it is | Rights |
|---|---|---|
| L01 | The verse, in Devanāgarī as printed | Free to everyone |
| L02 | The same verse in Roman letters | Free to everyone |
| L03 | Word by word | Published by Aesthetic Vedānta |
| L04 | Prose word orderReserved, and absent in this edition. The book supplies a word-by-word gloss, which is a different thing, and conflating the two would be a scholarly error. The number is held open rather than reused, so adding one later never renumbers a citation already made. An absent part is a valid state, not a fault. | None |
| L05 | Translation | Published by Aesthetic Vedānta |
| L06 | Commentary | Published by Aesthetic Vedānta |
| L90–L99 | Recitation audioReserved and unoccupied. The numbers were allocated before the roadmap reached them. | None |
What is registered
The whole registry, today
Read off the registry as this page renders, so none of it can drift from what the engine actually holds. The Mystery of the King of Narasiṁha Mantras, published by Aesthetic Vedānta (imprint of Aesthova Private Limited) in 2026, is the first and so far only registered work.
- Verses
- 13
- Parts
- 65
- Separately named objects
- 78
- Citations
- 66
- Works they reach
- 21
- Dictionary headwords
- 40
Where this strains
Two things a name cannot do
A seal proves the text is unaltered. It does not prove the text was right.
The cryptographic hash over a sealed book is a strong claim about one thing only: that nothing has changed since it was published. Whether the reading was correct when it was sealed, whether the locator was the right one, whether the translation renders the Sanskrit faithfully — none of that is a matter a checksum can settle. Those are scholarly claims, made by named people, defended in public, and revisable. The seal protects them from tampering; it does not make them true.
The dependency did not vanish. It moved.
It is tempting to say this removes reliance on anybody. It does not. It exchanges a dependency on an organisation that can be acquired, defunded or reorganised for a dependency on a mathematical problem that is merely hard. That is a very good trade and it is not the same as freedom. Hard problems occasionally become easy, and when one does, the honest response is to re-seal the corpus under something stronger — not to pretend the guarantee was unconditional.
And one thing this does not prove
None of this means the engineering proves the tradition was right all along. Building a durable naming scheme for a corpus says nothing about whether the corpus is true, and treating a working piece of software as evidence for a metaphysics would be exactly the kind of unearned claim this whole architecture exists to make harder.
What the tradition does supply is a diagnosis. The Gītā’s own account of its transmission says the chain was broken — naṣṭaḥ — and that is the strongest evidence in the neighbourhood precisely because it is an admission of failure rather than a claim of success. A text that reports its own line going dark is a text keeping records, and records are what a naming scheme is for.
The argument in full
The Address That Outlives the Server
The long version: how citation decay was measured, what Los Alamos found inside a single decade, what the Vedic corpus has been paying for considerably longer, and what it looks like once the repair is built in software.
Read the essay