3.4.6 — "in the Veda, [the affixes of]
लुङ्, लङ्, लिट् [may replace one another for any tense]." That is the whole rule, and
it does something a rule outside this domain never does: it opens a word,
छन्दसि, that the corpus's own inheritance graph
(C.sutra_rec()) shows being carried forward by अनुवृत्ति, unstated,
through every one of the next sūtras — 3.4.7 through
— before a sūtra finally drops it. Not one of those eleven
repeats the word. This page's own count of 132 is a count of sūtras that say the word
themselves; measured this way, it is a floor, not the true extent of the domain.
Two languages, one grammar
counted, not assumedछन्दसि is a licence, not a repair
3.4.6A sūtra that says छन्दसि is not fixing something broken. Classical Sanskrit is not an error the Veda made and Pāṇini corrected; the relationship runs the other way. The rule records that the older language did something the newer one no longer does, files it under one word, and moves on. Nobody is being corrected — the record is just being kept honest about which register a form belongs to.
A lakāra Classical Sanskrit does not have
3.4.7 – 3.4.17Page 11 (लकाराः) taught the ten lakāra a finite Sanskrit verb can be. It named an eleventh in passing — , shown only on a vowel grid, absent from every table of forms — and said no more, because that page is about Classical Sanskrit and this lakāra is not part of it. Here is the sūtra that licenses it, still inside the same 3.4.6 domain.
3.4.7 — wherever लिङ् (the optative) would be prescribed, an additional affix, लेट्, is available in the Veda alone, and the Kāśikā gives it in full:
C.krt()['lakara'] — the same vocabulary page 11 reads its
eleventh-lakāra note from — says outright: Nothing on this
page derives a लेट् form; every one shown is quoted from the Kāśikā's own citation
of the Ṛgveda and Saṃhitā texts, not generated.Classical Sanskrit has exactly one infinitive suffix, तुमुन् (3.3.10 ). 3.4.9 lists nine more that exist only inside this domain — से, सेन्, असे, असेन्, क्से, केन्, सेन्, ध्यै, अध्यै, न्कध्यै, कध्यै, न्शध्यै, शध्यै, न्तवै, तवेङ्, तवेन् — and the Kāśikā glosses several against the Classical word they stand in for:
Three more sūtras narrow further within the same domain: 3.4.13 gives तोसुन्/कसुन् for "capable of," 3.4.16 gives तोसुन् for a closed list of roots, and 3.4.17 gives कसुन् for two more. None of the four — से, तवै, तोसुन्, कसुन् — is available outside this one domain, and every one of them is superseded by तुमुन् the moment the domain closes.
Endings Classical Sanskrit retired
7.1.9 – 7.1.39Declension is where the two languages diverge loudest, because Classical Sanskrit spent several centuries regularising a case system the Veda still shows in flux.
7.1.9 is the Classical rule: after an अ-stem, the instrumental plural भिस् becomes ऐस् — वृक्षैः, प्लक्षैः, never *वृक्षभिः. 7.1.10 , inside the same domain, undoes it — variously:
Two directions at once, both in the same sūtra: the ऐस्-change reaches further than 7.1.9 allows it to (नद्यैः, from a stem that is not अ-final), and it also fails exactly where 7.1.9 says it must apply — देवेभिः and सर्वेभिः stand where a Classical text could only ever write देवैः and सर्वैः. बहुलम् earns its place in the very next section for exactly this reason: a rule that runs backward from its own default in both directions is not an exception, it is a description of usage that resists a single direction.
7.1.39 is the single densest sūtra in this domain — twelve replacements for the ordinary सुप् case-endings, each with its own Kāśikā citation set against what the ending "would have been" (प्राप्ते) in Classical usage. A sample of what it licenses, quoted directly:
Twelve substitutes in eleven words of sūtra, because Vedic declension is not Classical declension with a few endings swapped — it is closer to a family of endings the standard language chose one member from and let the rest lapse.
लुङ् without its augment
6.4.71 – 6.4.75Page 11 built अभवत् (लङ्) by adding, on top of the ordinary शप्-machinery, one augment: अट्, placed by 6.4.71 on every लुङ्/लङ्/लृङ् form without exception in Classical Sanskrit. The Veda has a form with the augment removed, and it is not a mistake — it is a category grammarians call the injunctive, and Pāṇini reaches it in three steps inside the same six-sūtra span.
6.4.71 is the rule page 11 used: अट्/आट् on every लुङ्/लङ्/लृङ् form, no condition attached. 6.4.74 narrows it — after the prohibitive particle मा, the augment does not appear at all. This part is ordinary Classical Sanskrit, not Vedic: मा कार्षीत् ("let him not do it"), never *मा अकार्षीत्.
6.4.75 , inside the same domain opened at 6.4.72, states the Vedic picture directly:
Without मा — where Classical grammar requires the augment — the Veda variously omits it anyway: an augment-less aorist used as a bare narrated fact, which is exactly what the term injunctive names. And with मा — where Classical grammar forbids the augment — the Veda variously keeps it, the mirror case. बहुलम् names both directions in one word, because no single direction is the rule.
बहुलम् — the most honest word in the grammar
3.1.85 · 3.3.1Every other word on this page marks a boundary: this is Vedic, that is Classical. बहुलम् — "variously," "in multiple ways" — marks the place where Pāṇini looked at the boundary and could not draw it cleanly. It is not a synonym for छन्दसि. It is what he writes when the data — Vedic or not — resists a rule that always holds, and he says so rather than forcing one.
3.1.85 — "interchange [of the विकरण affixes], variously." Its own eight syllables never say छन्दसि; they do not have to, because the word is still open from 3.1.84 above it. The Kāśikā says the domain outright, in its very first clause:
3.3.1 opens the entire uṇādi
appendix — 748 sūtras of affixes this series' own page (uṇādi is not yet a page
of its own; see C.unadi()) accounts for vocabulary the regular kṛt
rules cannot reach — with the same word, and no mention of the Veda whatsoever:
बहुलम् here means that an उणादि affix is not confined to the root it was coined for; usage decides, and the appendix exists because the regular rules could not. Two more, briefer, from elsewhere in the grammar, neither about Vedic at all:
Two more places the two languages split
already taughtClassical Sanskrit fixes a गति or उपसर्ग immediately before its verb (1.4.80). 1.4.81–82, inside this same 132-sūtra count, let the Veda put it after the verb, or separate it by an entire clause. Page 15 (उपसर्गाः) already walks this with the Kāśikā's own Ṛgveda quotation — नि याति versus नियाति, and the Indra invocation split across a full verse — so it is not repeated here.
Page 04 (स्वरः) counted it already: not one of this corpus's 3,983 sūtras, 9,007 paradigms or Devanāgarī root-forms carries an accent mark. That absence is exactly why प्रग्रह, स्वरितत्व and the whole apparatus of Vedic pitch accent cannot be taught from this site with worked examples — the record that would let a reader check a prediction does not survive here.
One nuance belongs on this page and nowhere else: of the 132 sūtras counted above, have Kāśikā commentary that does carry an accent mark — ॑ udātta, ॒ anudātta, the exact two marks page 04's own count checked for — because the commentary is quoting real Saṃhitā lines, not writing unaccented prose. जोषि॑ष॒त् above still carries its own accent, untouched, inside the Kāśikā that cites it. The gap page 04 reports is real and it is specific: the sūtras and paradigm tables carry no accent; the Vedic quotations embedded in commentary, where they survive at all, still do.
What this buys a reader
practicalA form that no rule in adhyāya 1–7's ordinary machinery generates is not automatically a mistake, in your own writing or in someone else's. Before deciding a form is wrong, check whether one of these 132 sūtras — or one of the further sūtras that inherit their word by अनुवृत्ति without repeating it — licenses exactly this case. जोषिषत् looks like a broken लिङ्; it is a correctly formed लेट्. देवेभिः looks like a bhis that forgot to become ऐस्; it is what 7.1.10 says the Veda actually writes. That is not a loophole in the grammar. It is the grammar telling you, in advance, which forms belong to which register.
What this page cannot check
honestEverything Vedic quoted above comes from one
place: the Kāśikā's own citation of the sūtra it is explaining, pulled with
C.layer('kashika', ref). This corpus ships no independent Vedic corpus to
check a form against, and vidyut-prakriya — the engine pages 11 and 12 read
live derivations from — is documented, in its own data, as having poor support for
लेट्, the one lakāra this page's central example turns on. No form on this page was
generated. Every one is a quotation, and every quotation above is checked at build time
against the exact corpus string it claims to come from — the assertions are in
A12-chandasa.data.py, not asserted here in prose.
All 132, by marker and adhyāya
InteractiveEvery sūtra this page's own scan of C.index_rows() found —
filter by which word licenses it, or by which adhyāya it falls in. Kāśikā gloss shown
where the sūtra draws one.
Where the 132 fall
ReferenceBy adhyāya, and by what kind of rule each sūtra is — संज्ञा, परिभाषा, विधि, अतिदेश, अधिकार — the same five-way split page A8 used for the वार्त्तिक.
| अध्यायः | सूत्राणि | चाान्दस | % |
|---|
| प्रकारः | सूत्राणि | चाान्दस | % |
|---|
C.index_rows() gives all 3,983 sūtras with their own text and rule type; a
sandhi-aware regular expression (छन्दसि, मन्त्रे, वेदे, बहुल[ंम्]?, ब्राह्मणे — the
anusvāra spelling of बहुलम् included, or eleven of the 132 would have gone uncounted)
scans every one of them. C.sutra_rec() gives the अनुवृत्ति chain in
§2 and confirms it stops exactly where claimed. C.layer('kashika', ref)
supplies every Vedic form and every gloss quoted above; nothing on this page was typed
from memory, and every quotation is checked against that same string at build time.Check yourself
अभ्यासःSix questions.