On Aṃśa: a letter to a colleague
A letter to a colleague, 31 July 2026 — on the name Aṃśa, and on the forms that keep failing it.
Your note said, roughly: surely the marks are a nicety. Permit me to use Aṃśa to show they are not.
First, the attestation. The attested form is अंश, from the tradition's own record. The canonical transliteration is Aṃśa.
And the marks themselves: s → ś (Stress mark (acute/tone)). Remove any one of them and you have not simplified the name — you have told a small lie about it, and references are built of exactly such small truths.
Say it aloud as the sources heard it: uM-shah; the ṃ is a nasal hum, the ś a soft sh. The marks are what carry that sound to a reader who has never heard it spoken.
On the plain form — and I want to be precise, because you pushed on this. The plain form amsa is not false — it is a fallback, and the register is exact about what that means. In systems that cannot carry diacritics (legacy databases, older tooling, parts of the address-bar pipeline) the flattened form is the accepted vehicle, and the rulebook names it explicitly as the final rung of the fallback hierarchy: full restoration first, circumflex where attested, macron-only by the academic standard, an attested variant, and only then plain ASCII.
What the fallback status forbids is promotion: amsa may carry the name where marks are impossible, but it may not represent the name where marks are possible. Every mark it drops — s → ś — is a fact the sources recorded. A fallback is a vehicle, never the primary.
Now to the forms that provoked your letter. As for Amsa: it comes from the publisher compromise — older books and hurried websites that print the "English equivalent" as if it were the name itself, and it fails because every substitution here (ṃ→m, ś→s) is a romanization of convenience, not an attested spelling — it teaches the reader to guess, not to read
One more thing, because it always surprises: the name did not stay home. He overlaps with Bhaga, the older "dispenser" of shares — the two are so close that Vedic lists sometimes trade their places — and with Mitra and Varuṇa, whose covenant-keeping the Ādityas collectively perform. Scholars compare the Indo-Iranian share-gods: Avestan aṃša (in Amesha Spenta, "Bounteous Immortal") is a cognate word, if a different theology — the portion raised to archangel.
And it never stopped working. His word outlived his cult: aṃśa remains the everyday Sanskrit and Hindi word for "part," and technical vocabularies are built on it — navāṃśa (the ninth harmonic chart in Jyotisha astrology), kalāṃśa in music, daśāṃśa in taxation records. Wherever Indian thought divides a whole into rightful portions — astrology, inheritance law, music theory — the god's name is the operative term.
The Convention's Own History
Long before IAST was formalized in 1912, Sanskrit had Pāṇini — the fourth-century-BCE grammarian whose rules describe the language's sounds with a precision no other ancient tradition matched. That heritage is why the convention is so exact: retroflexes (ṭ, ḍ, ṇ, ṣ) are not accented dentals but separate phonemes with separate meanings; long and short vowels (ā/a, ī/i, ū/u) change words, not just timing. IAST writes those distinctions down with underdots and macrons so that a reader who has never heard Sanskrit can still read it correctly. Dropping the marks is not transliteration at a lower resolution — it is a different, lesser claim.
Where Protocol Meets Practice
The technical answer is simple: the canonical marks are legal IDNA 2008 characters, encoded as punycode under `xn--`. The practical answer is messier: legal does not mean stocked (registrar inventory is partial), and stocked does not mean typeable (stacked diacritics defeat mobile keyboards).
So the register's rule is a hierarchy of attested forms — full restoration, circumflex, macron-only, attested variant, and finally plain ASCII for the systems that cannot carry marks. Every rung is defensible because every rung is attested. The forms this register rules against are not rungs; they are errors, and errors do not get rungs.
Write back when you have checked the sources; they are better company than I am.
— The Register
The Verdict
Aṃśa is the canonical form: attested, checkable, and enforced across the temple, the lexicon, the cards, the APIs, and this register. The plain form amsa is sanctioned as the fallback vehicle and nothing more. The false forms above are documented so that no reader has to take the temple's word on faith — the violations are stated, the sources are named, and the register stands open to challenge on every line.
The temple of Aṃśa stands at /sites/amsa/.
The docket for this ruling: 1 false form ruled against, Amsa first on the docket; 1 preserved mark verified: s → ś; no contested forms attested; the attestation rests on 6 cited sources — Rigveda, Brāhmaṇas, Bhagavata.
The Register Continues
The register holds 287 entries like this one — each temple with its own ruling. Read the temple's founding dispatch for the story, its Restoration File for the philology, its Resonance File for the industries, or the whole archive end to end.
Continue the register — previous entry: Amitābha (read it) · next entry: Ꜣmun (read it).
The Canonical Register is written from the same canonical record as the lexicon itself. If a false form is missing from this page, it is because the evidence does not support calling it false — or because we have not seen it yet. Show us, with a citation, and the register will answer.