A sūtra is never the whole rule — three earlier pages already showed why. But before any of that helps, one question comes first, and it is the one beginners skip: what kind of statement is this even making? A definition reads differently from an operation, and an operation reads differently from a heading that never operates on anything at all. Tell the six apart and the 3,983 stop being an undifferentiated wall of Sanskrit — each one tells you, in its own shape, what to do with it.
Read this cold, the way it sits on the page: . Four words. Nothing in them says what a root is, what an affix is, or why a letter like ङ् matters to any of it. It is not a hard sentence — it is an incomplete one, and it is incomplete on purpose.
Pāṇini wrote 3,983 sūtras short enough to memorise as a set because almost nothing in any one of them is written twice. What is not written is recovered four ways, and this page names all four before handing each one to the page built to teach it in full: अनुबन्ध and प्रत्याहार (page 01), अनुवृत्ति (page A2), अधिकार (page A3). Below, this same sūtra gets all four unpacked at once, so you can see exactly how much of it was never on the page.
Every sūtra in the corpus carries a type — not this page's
judgement, but a field annotated against the tradition's own commentary. Five plain kinds,
and a sixth pattern worth watching for: some sūtras are two kinds in the same breath.
of the 3,983 — . A विधि changes something: inserts, replaces, elides, or forbids. — 6.1.77 — is one: something (इक्) becomes something else (यण्) under a condition (अचि). Once the case roles are read, every विधि has that verb-shaped feel: a change, happening to something, under a trigger.
of the 3,983 — . A संज्ञा names a technical term so later sūtras can use one word instead of a description. — 1.1.1 — is the very first sūtra of the text and one of the shortest: two words, "आत् and the ऐच्-sounds are called वृद्धिः." Nothing happens; a name is minted.
of the 3,983 — . An अधिकार states a range and supplies one word to every sūtra inside it, silently, without doing anything on its own. — 6.4.1 — is two syllables that govern six hundred and thirteen later sūtras. Page A3 covers this in full; here it only needs a name.
of the 3,983 — . An अतिदेश borrows rules written for one thing and extends them to another by declared analogy. — 1.1.56 — says a replacement (आदेशः) behaves like the thing it replaced (स्थानिवत्) for every purpose except rules about sound itself. The telltale word is usually वत् — "as if."
of the 3,983 — , the rarest kind. A परिभाषा does not describe Sanskrit; it tells you how to read the other four. — 1.1.49 — is one of six such reading instructions this page uses throughout; see them all below.
One विधि, four words, twenty-seven syllables — and it happens to use every compression device the grammar has. Each is taught properly elsewhere; here they meet in one sūtra so you can see all four at once.
— genitive plural, "of those not इदित्": roots that do not carry इ as their own it-marker in the dhātupāṭha. And — locative, "when [the affix] is क्ङित्": marked with क् or ङ् as its it-marker. Neither क् nor ङ् nor the root-level इ ever appears in the derived word — each is a coordinate stuck onto a citation form so one letter can pick out a whole class. Full mechanism on page 01.
— genitive, "of a हल्": any consonant. हल् is built the same way as अच् and इक् on page 01 — a span of the fourteen माहेश्वर sūtras, closed on an इत् — and this page's own corpus record for it lists members (it lists ह् at both ends of the span, page 01's own caveat). First fixed as a coordinate at .
6.4.24 states no operation of its own. — 6.4.23 — states it: elide the न्. 6.4.24 inherits that word, नलोपः, one hop back, by sense alone — nothing marks the inheritance in 6.4.24's own eight syllables. Full graph, all 9,266 edges, on page A2.
(6.4.1) has been open for twenty-three sūtras and stays open for five hundred ninety more: this is about a stem. (6.4.22) opened two sūtras ago: evaluate this as if the other rules in this zone had not yet fired. All 73 domains, mapped and measured, on page A3.
Put back together, step by step:
Page 03 taught this for सन्धि specifically: a rule names its roles by case alone, not by word order or by any label. षष्ठी marks what is replaced, प्रथमा the replacement, सप्तमी the trigger, पञ्चमी what precedes. That was never a सन्धि-only trick — it is how every विधि in the grammar is built, and six sūtras state the reading instructions that make it exact.
| Case | Marks | Reading instruction |
|---|
Two of them refine the basic four rather than adding new roles: a षष्ठी-marked replacement does not swallow the whole word it names — only its last sound — and a पञ्चमी-marked trigger does not affect the whole word that follows — only its first. A sixth, 1.1.72, explains why a rule can name a single sound and still reach every word that ends in it.
Every reading above collapses into one habit. Run it on a sūtra you have never seen, in this order, before trying to translate a single word of it.
Only after all four does what feeds off it — usedBy — matter,
and only if you are tracing a rule's consequences forward rather than reading it. The
picker below runs the first three steps for you on any sūtra in its set, so you can spend
your attention on the fourth.
Search by Sanskrit text, by a romanised guess at it, or type a reference
directly. Diacritics are folded, so vrddhi finds वृद्धि
the same as वृद्धि does.
The full corpus count for every kind, exactly as tallied above.
| Code | Devanāgarī | Does | Count | Share |
|---|
C.index_rows()
gives every sūtra's type; the five counts and the eighty-five mixed sūtras
are walked fresh from it and from the raw sutra/*.json shards at build time,
never hand-typed. Shipping full detail for all 3,983 would cost more than this page's size
budget on the raw Devanāgarī alone, so the picker above ships a curated slice
instead: every one of the 73 अधिकार, 67 अतिदेश and 24 परिभाषा sūtras — cheap, and the three
rarest kinds — plus a sample of संज्ञा and विधि stratified across every pāda that has any,
so a search rarely comes up empty no matter which part of the grammar it touches. Anything
a sūtra in the picker inherits from outside that slice is still shown by its real text,
just without its own padaccheda to expand further — the reading view
(Grammar/site/read.html) has the rest.Six questions on the diagnostic itself — not on any one sūtra's meaning.