How a folio gets made
Machines do the volume. People hold the authority. Nothing reaches a reader without a named human standing between the model and the mantra, and where no human has yet stood there, the page says so in the reader's face rather than in a footnote.
The six steps
Acquire
Public domain editions, manuscript scans and institute archives. The physical source and its edition are recorded before a single character is read.
If we cannot say which edition a line came from, the line does not go up.
Break to the smallest unit
Maṇḍala, sūkta, ṛc, pada, so that any quotation anywhere on the internet can point back at exactly one line, permanently.
One verse, one address, forever.
Cross-check
Every unit is set against a second independent digital edition. Where two editions agree, we say so. Where they differ, we print both readings and name which source gave which.
Agreement between independent witnesses is evidence. A single source is only a claim.
Carry across
Each language is drafted against the padapāṭha and the commentaries, never against another translation. English is not the pivot, which is why the Tamil here is not a translation of Griffith.
Terminology is locked to a shared glossary before any drafting begins.
Sign
A named scholar or reciter signs each unit. Their name, school and date sit on the folio. Unsigned material stays marked as draft in the reader’s face, not in a footnote.
The signature is the product. Everything else is typesetting.
Give it away
Every verse is a stable URL, a JSON response and a bulk download. A library that cannot be copied out of is not a library, it is a collection.
Corrections welcome, credited by name, history public.
What the labels mean
Three labels appear on verse pages, and they are the most important things on them. Reading them correctly matters more than reading the translations.
Signed
A named human translator and a citable edition exist for this rendering. The name and the edition are printed beside it. You can go and check. Most signed translations in the library right now are public domain works from the nineteenth century, which means they are reliable as scholarship of their period and carry the assumptions of their period with them.
Unsigned draft
Nobody has put their name to this. It is shown because a reader with no Sanskrit is better served by a labelled draft than by a blank space, but it carries no authority and should not be quoted as if it did. Every draft has a link next to it for a qualified reader to correct or sign it.
Agreement
Every unit is set against a second independent digital edition. Three outcomes are possible and all three are printed:
- Confirmed, two independent editions agree on the reading.
- Single source, only one edition was available, so the reading is a claim rather than a corroborated fact.
- Divergent, the editions disagree. Both readings are printed with the source of each named. We do not pick a winner silently.
On the read aloud button
Machine speech is a fourth category, and it gets labelled like the rest.
There is no Sanskrit voice on any operating system. Asked to speak Devanagari, a browser reaches for a Hindi voice, which applies Hindi phonology to Vedic Sanskrit: it deletes schwas that should be sounded, gets vowel lengths wrong, and ignores the udātta and anudātta marks completely. Those marks are not decoration. Moving an accent can invert a meaning, which is the whole point of the classical exampleindraśatru, where the accent decides who kills whom.
So the read aloud control is a pronunciation aid for someone who cannot read Devanagari, and it says so every time it is pointed at Sanskrit. It never plays the Sanskrit by default, it names the borrowed voice your device chose, and it is never called recitation. It also refuses to read unsigned drafts aloud, because a spoken sentence carries a confidence that a draft has not earned.
Recitation means a named human in an unbroken paramparā, with the śākhā and the teacher recorded. Where such a recording exists it plays instead, with the reciter named beside it. Where none exists the page says so plainly, because that absence is the most time-critical gap in this entire project. About eleven branches are still recited. Every one of them is a person who will not always be here.
What we will not do
- Put up a line without being able to say which edition it came from.
- Reconstruct, smooth over or fill in a damaged or missing passage.
- Present a machine rendering as though a person had checked it.
- Call synthesised speech recitation, or let it play the Sanskrit unasked.
- Translate one translation into another. Each language is drafted against the padapāṭha and the commentaries, which is why the Tamil here is not a translation of Griffith.
- Hide a gap. Where nothing survives, the page says nothing survives.
Corrections
Errors are ours and get fixed in the open. Corrections are credited by name and the history stays public. If you find something wrong, the fastest route is the contribution page.