Kṛṣṇa wants a son. He goes to the sage Upamanyu, undertakes austerities, and Śiva grants him one, called Sāmba.
The volume narrates that twice, a hundred and eight chapters apart, in two different Saṃhitās. The two versions disagree about where he went, what initiation he received, how long it took, how many boons he got, and who the child actually is.
Then an endnote supplies a third identity that matches neither.
⚑ But the doublet is not the finding here. This volume has several. What makes this one worth a page is a single clause in the second version: the sages open by saying "We have heard this earlier" — and what they are referring back to is not the first version. It is a different account, in their own Saṃhitā, involving a character the first version does not contain.
The Vāyavīya cross-references itself correctly and does not know the Umā Saṃhitā's telling exists. That is evidence about how this book was assembled drawn from the text's own referential behaviour, which is a much harder kind of evidence than style.
| ch 315–317-6(1–3), Umā | ch 423-8.2(1), Vāyavīya Uttara | |
|---|---|---|
| Narrator | Sanatkumāra → Vyāsa | Vāyu → the sages at Naimiṣāraṇya |
| Where he goes | "Krishna, Vasudeva's son, went to Kailasa, Shankara's abode" and finds Upamanyu there | Upamanyu's hermitage |
| The initiation | "the sage performed his diksha. He taught him Sharva's great Atharvashirsha mantra" | "smeared his body with bhasma… jnana about how one should perform a Pashupata vrata, lasting for twelve months" |
| How long | Śiva appears "In the sixteenth month" | "At the end of a year" |
| Boons | eight, itemised — plus a second set from Pārvatī | none |
| Who the son is | *"The sages cursed the Samvartaka sun that he would be born as a human. He is the one who will become your son."* | "Jambavati's son got the name of Samba" |
| Why "Sāmba" | not explained | "When Mahadeva gave him this son, he was with Amba. Therefore… Samba" |
12
⚑ Note that the disagreements are not degrees of detail. Kailāsa against a hermitage; an Atharvaśiras dīkṣā against a twelve-month Pāśupata vrata; sixteen months against twelve; eight boons and a second set against none. And the son is a cursed cosmic body in one and a wife's child in the other.
⛔ Unflagged. The only note attaching to the identity is 1918, which does something else entirely.3
The disagreements are easier to notice, so the agreements are worth setting out — they are what makes this one episode rather than two.
Kṛṣṇa wants a son. He goes to Upamanyu. Upamanyu is a Śaiva ascetic with an austerity history of his own. Kṛṣṇa undertakes a Śaiva discipline under his instruction. Śiva appears. The son is granted and is called Sāmba.
⚑ That is a complete narrative skeleton and not one joint of it moves. What moves is everything technical hung on it: the location, the initiation, the mantra, the duration, the boon count, the child's identity.
⚑⚑ A stable skeleton with divergent technical furniture is the signature of two traditions telling one story. A single tradition mis-remembering itself corrupts the story — names drift, order slips, episodes merge. Here the story is intact and the specifics are wholesale different, which is what you get when two communities each narrate the same received event in their own vocabulary.
⛔ That is a reading of one episode and is marked as one. What supports it beyond the episode is the referential evidence below, which is independent of it.
One asymmetry deserves separating out, because it is not a difference in detail — it is a difference in what kind of scene this is.
The Umā telling is a boon scene. Śiva appears and grants eight itemised boons, and Pārvatī then grants a further set of her own.1 The son arrives inside that transaction, as one item among others, with an aetiology attached.
The Vāyavīya telling grants nothing but the son. Śiva arrives "accompanied by Amba and the foremost ganas", is "pleased with the austerities", and gives a child.2 There is no list, no negotiation, no second grantor.
⚑ And the naming follows the difference exactly. In the Vāyavīya, Ambā's presence at the moment of the grant is what the child is named for — "When Mahadeva gave him this son, he was with Amba" — which only works in a scene where the grant is a single, specific event. In the Umā telling, where Pārvatī grants a whole separate list, that etymology has nothing to attach to, and the text does not attempt one.
⚑⚑ So the two versions are not just differently detailed. Each one's naming logic is coherent with its own scene structure and incoherent with the other's — which is a tighter fit than a compressor or an expander would produce, and the strongest internal reason to treat them as two compositions rather than one edited twice.
The second telling opens with a request, and the request contains a claim about what the audience already knows.
"O illustrious one! You have already narrated the great-souled Upamanyu's account… Vasudeva Krishna, unblemished in deeds, went and met Dhoumya's elder brother and performed Pashupata vrata. He obtained supreme jnana. We have heard this earlier. How did Krishna obtain the supreme jnana about Pashupata?"4
Three things in that sentence are checkable, and all three check.
One — Dhaumya. A whole-volume sweep puts the name at exactly five places: lines 1275, 1279, 1291, 1305 and 1311. Every one is in the Vāyavīya.5 The Umā Saṃhitā's version, ch 315–317, does not contain him.
Two — the Pāśupata vrata. The word does not occur anywhere in ch 315–317.5 That version's initiation is a dīkṣā with the Atharvaśiras mantra.
Three — "We have heard this earlier." ⚑ They have. The Vāyavīya narrated Upamanyu's own story in its own preceding chapters, and the back-reference lands on it correctly.
⚑⚑ So the Vāyavīya is not being careless. It is being precise about a different text. Its cross-reference works, its terminology is internally consistent, and the account it points back to is its own. What it does not do is register that the same episode was narrated a hundred and eight chapters earlier, in the same printed book, with a different initiation and a different son.
⛔ Stated at its real strength: this is one episode with one working cross-reference, and one instance does not establish a compositional theory on its own. What it does is put the claim on a different footing from a stylistic impression — the text is demonstrably tracking its own references, and its tracking stops at a Saṃhitā boundary. [n=1 OBSERVED] on this episode; the conclusion it supports is argued alongside independent evidence elsewhere.6
Note 1918 attaches to the naming clause in the Vāyavīya version:
*"Samba = Sa (with) + Amba. Jambavati was one of Krishna's wives. Guha was born as Samba."*3
Guha is Kārttikeya, Śiva's own son. The note asserts, unhedged, that the child is a rebirth of him.
⛔ That matches neither body version. Ch 317 says the child is the cursed Saṃvartaka sun; ch 423 says he is Jāmbavatī's son named for Ambā's presence. The apparatus supplies a third identity in the same typeface, in a single sentence, with no marker that it is coming from outside.
⚑ And Guha is not a stable name in this volume either. At ch 311-5(40), Śiva renames the hunter Gurudruha: "Pleased with him, Shiva gave him the name of Guha", and places him at Śṛṅgaverapura — which is the Niṣāda Guha of the Rāmāyaṇa, not Skanda.7 Two Guhas, one volume, no note.
⚑ Which makes the note's own reference ambiguous inside the book it annotates. It is glossing a name whose other occurrence in the same volume is a different person.
⚑ Line them up and what is striking is that each is complete. None is a hedge, a variant reading, or a partial gloss.
⛔ Nothing in the printed book distinguishes their status. The first two are body text and the third is a note, but the note carries no hedge, and this apparatus hedges freely elsewhere — forty-eight notes contain probably, presumably or seems.8
⚑ So a reader who wants to know who Sāmba is in the Shiva Purāṇa gets three answers from one printed volume, and the one that sounds most authoritative — because it is shortest and reads as a definition — is the one that is not in the text at all.
You have found the same story twice in one corpus and want to know whether it is a variant or a join.
Test the cross-references, not the prose. ⚑ Style comparison is soft; reference behaviour is hard. Take every "as we have said", "described earlier", "you have already narrated" in each version and check where it lands. A version whose back-references all resolve inside its own section, and which never points at the other version, is not a variant of it — it is a body of material that was working from something else.
Watch for a back-reference that lands on the wrong thing. ⛔ Ch 423's sages say "we have heard this earlier" about a Pāśupata vrata and a character who exist only on their side of the book. That is not a false claim — it is a true claim about a different corpus, which is a much more informative failure than a simple error.
Count what changed and what did not. ⚑ Here the episode is stable — Kṛṣṇa, Upamanyu, austerities, a son from Śiva — and everything technical is different: the place, the initiation, the interval, the boon count, the child's identity. A stable narrative skeleton with divergent technical furniture is the signature of two traditions telling one story, not of one tradition misremembering.
And read the notes as a separate witness. ⛔ The third identity here is the editor's, is unhedged, and sits in the same typeface. When you inventory what a source says, inventory what the apparatus says separately — otherwise you will attribute to a text a claim no manuscript of it contains.
Evidence. Both tellings read in full in place: ch 315-6(1) through 317-6(3), lines 253–285 — the journey to Kailāsa, the dīkṣā and Atharvaśiras mantra, the sixteenth-month prediction, the eight boons and Pārvatī's second set, and the Saṃvartaka-sun identification;1 ch 423-8.2(1), lines 1299–1311 — the sages' back-reference, the bhasma and twelve-month Pāśupata vrata, the year-end grant, and the Sa+Ambā etymology.2 Note 1918 read in full.3 Scripted whole-volume sweeps for Dhaumya and Pāśupata across both ranges.5 Ch 311-5(40), line 207 range, for the second Guha.7
⛔ Cited in the Saṃhitā-internal form throughout — the running chapter numbers in this edition are not unique, and neither hazard is stated anywhere in it.9
Tension 1 — the two versions are not equally weighted in the volume. ⚑ The Umā telling runs three chapters and the Vāyavīya's one, and the Umā version's boon list is far longer. A compression reading is available: the Vāyavīya may be summarising a known story rather than telling a rival one. ⛔ What that reading does not explain is the changed initiation and the changed child — a summary drops detail, it does not substitute a different vrata and a different mother.
Tension 2 — the Umā Saṃhitā is the volume's expansion zone. It runs far over the shloka budget the text publishes for itself, and it is where every imported frame this build located sits.6 ⚑ So the more likely candidate for the late arrival is the Umā version rather than the Vāyavīya's, which is the opposite of what chapter order suggests. [PLAUSIBLE — needs corroboration]; nothing here decides it.
Tension 3 — the third identity may be tradition rather than invention. ⛔ Note 1918's "Guha was born as Samba" is very likely drawn from somewhere in the wider Purāṇic literature. This build has not located a source for it and does not claim it is fabricated. What is claimed is narrower and holds regardless: it is not in this text, it is unhedged, and nothing marks it as external.
Open question. The back-reference test worked once. Run it across every "as narrated earlier" in the volume: how many resolve inside their own Saṃhitā, how many cross a boundary, and does any Umā back-reference land in the Vāyavīya?
Against The Chakra, Given Twice: the volume's other narrative doublet, and the contrast is the useful part. ⚑ There the two tellings sit two chapters apart with a connective sentence between them saying after this. Here they sit a hundred and eight chapters and two Saṃhitās apart, each with its own working cross-references. Same phenomenon, opposite distances, and neither explains the other — which is why they are two pages.
Against The Kailāsa Is a Different Book: adjacent evidence for the same conclusion by a different route. ⚑ That page's markers are technical — tattva counts, Praṇava components, cited authorities. This one's is referential: a Saṃhitā tracking its own material and not the neighbour's. A technical divergence and a referential blind spot are independent instruments, and they agree.
Against The Saṃhitā That Grew: the awkward direction. ⚑ That page's finding is that the Umā runs far over its published budget and carries the imported frames — which points at the Umā telling as the late one, not the Vāyavīya's, even though it comes first in the printed order. ⛔ This page records that tension rather than resolving it.
Against The Apparatus Is the Translator's: note 1918 is a clean specimen. ⚑ A one-sentence identification, unhedged, in the same typeface, supplying a third answer to a question the body answers twice — and the reader has no way to tell it apart from the text.
Against Only What a Character Asks: this doublet is out of reach of the volume's detection mechanism by construction. ⚑ Detection fires when a listener raises an objection, and listeners see only as far back as the conversation they are in. The sages at Naimiṣāraṇya are in the Vāyavīya's frame; Vyāsa heard the other version from Sanatkumāra in a different frame entirely. Nobody is in a position to have heard both.
Book Killing Book — Textual Genealogy — that page's territory is how texts descend from and displace one another, and how the line of descent is reconstructed.
The Purāṇic case contributes a specific and unusually clean instrument, and the contribution is methodological rather than substantive.
Genealogy is normally argued from shared error, shared wording, or shared omission — the classic moves of stemmatics. ⚑ This episode adds a fourth: the reference that resolves correctly on one side of a join and finds nothing on the other. Ch 423's sages say "we have heard this earlier" about a Pāśupata vrata and Dhaumya's brother, and both exist — in the Vāyavīya only. The claim is true and its truth is bounded by a section break.
That is a stronger signal than shared wording, for a reason worth stating: wording can be copied by anyone with the manuscript in front of them, but a working cross-reference records what its author had actually read. A compiler stitching two traditions together can harmonise vocabulary in an afternoon. Harmonising every back-reference means auditing what each section assumes its audience already knows, which is expensive, invisible if done, and glaring if skipped.
⚑ So the diagnostic generalises: in any composite document, the internal pointers are the last thing anyone updates. A merged codebase, a revised standard, a policy assembled from two predecessors — the prose gets reconciled and the cross-references keep pointing home. ⛔ Which also supplies the caution: a pointer that resolves proves contact, not priority. Ch 423 knows the Vāyavīya and not the Umā; that establishes a boundary, and it says nothing about which side is older — and this page, and the volume's own budget evidence, disagree about that.
Sharpest implication. Kṛṣṇa goes to Upamanyu for a son, and this volume tells it twice. In the Umā he "went to Kailasa, Shankara's abode", receives a dīkṣā and "Sharva's great Atharvashirsha mantra", waits "In the sixteenth month", and gets eight boons plus a second set from Pārvatī — and the son is "the Samvartaka sun", cursed by sages into human birth. In the Vāyavīya he goes to a hermitage, is smeared with bhasma, performs a "Pashupata vrata, lasting for twelve months", waits "at the end of a year", receives no boons, and the son is "Jambavati's son", named because Śiva "was with Amba." ⚑⚑ And the second version opens with a back-reference that is true about a text this one is not: *"Vasudeva Krishna… went and met Dhoumya's elder brother and performed Pashupata vrata… We have heard this earlier."* Dhaumya occurs five times in the volume and all five are in the Vāyavīya; the word Pāśupata does not occur in ch 315–317 at all. The Vāyavīya cross-references itself correctly and does not know the Umā Saṃhitā's telling exists. Then note 1918 arrives, unhedged, in the same typeface: *"Guha was born as Samba."* Neither body version says that — and Guha is also, sixteen chapters earlier in the same volume, the name Śiva gives a hunter. Three complete identities for one child, from one printed book, and the one that reads most like a definition is the one no manuscript carries.
Generative questions.