Author — Anonymous (traditionally ascribed to Krishna Dvaipayana Vedavyasa); translated by Bibek Debroy
Year — Text consolidated c. 1000–1400 CE; Debroy narrows to 1000–1100. This translation: Penguin Classics, the third and final volume.
Original file — RAW/books/Shiva Purana Volume 3 - Bibek Debroy.md (converted from EPUB 2026-08-11 by a spine-ordered script, diacritics verified intact; 5,915 lines, 247,200 words. Body at lines 129–1599 — 1,471 lines, 218,308 words, 160 chapters. Endnote layer at lines 1600–5860 — 4,261 lines, 22,195 words, 2,124 parsed notes.)
Source type — Primary text in unabridged scholarly translation, with a scholarly apparatus that is itself an object of study here
Publisher — Penguin Books (Penguin Classics)
Sanskrit source text — Sivamahapuranam, Nag Publishers, Delhi, 1986
Coverage — 160 chapters, ch 305–463, in four sections: Koti Rudra Samhita concluded (ch 305–314); Uma Samhita (ch 315–365, i.e. 6(1)–6(51)); Kailasa Samhita (ch 366–388, i.e. 7(1)–7(23)); Vayaviya Samhita, Purva Bhaga (ch 389–422, 8.1) and Uttara Bhaga (ch 423–463, 8.2). This volume completes the Purana. Its last line is "This ends Shiva Purana."
Endnote layer — Introduction notes 1–14; body notes then restart per section. 2,124 notes parsed in total, all censused; every note matching an intervention marker read in full.
Volume 3 is the doctrinal and ritual half of the Shiva Purana, and it is where the text stops narrating and starts specifying. The Uma Samhita runs sin, the hells, Yama's court, the body, death-omens, the ancestors and the Devi cycle. The Kailasa Samhita is not Purana at all but a tantric ritual manual on Pranava — mandalas measured to the square, bija mantras given by alphabetic position, six chakras, nyasa, sannyasa-diksha in full procedure. The Vayaviya Samhita, the largest section at 76 chapters, retells the Daksha catastrophe a third time and then builds a complete Shaiva institution: initiation, the guru, samskaras, the fivefold rite calendar, lingam installation, yoga, and the disposal of the dead.
The argument under all of it is that the text has an exhaustive account of desert and a total amnesty from it, running side by side, in nineteen separate chapters, on at least five currencies, without ever placing them in one argument — and that when it does reconcile them it does so five separate ways and never notices it has done so more than once.
The second argument is not the text's but the edition's. Debroy's apparatus performs roughly twenty-three disclosed editorial acts in this volume, each disclosed once, in an endnote, in a numbering scheme that is not unique — and every one of them moves the printed text toward the customary, the consensual, the internally consistent or the natural-sounding. That finding is the reason this stub carries a longer Limitations section than either of its predecessors.
Classification rationale. Filed as primary-text for the same reason as Volumes 1 and 2 — the object of study is the Purana. The difference in this volume is that the apparatus has itself become an object of study, and where a page cites the apparatus it says so.
⛔ TWO CITATION HAZARDS, NEITHER STATED ANYWHERE IN THE EDITION.
388-7(23) and the Vayaviya's first 388-8.1(1); and 365 is never used. 158 distinct numbers plus 2 duplicates = 160 headings, and the running total closes at 463 against a stated 464 — short by exactly one, which is what two duplicates and a gap produce. The Samhita-internal numbering is exact throughout. ⛔ Cite this edition as 388-8.1(1), never as "ch 388." The running number is the translator's own, by his statement in the Introduction, and the failure is unflagged in all 2,124 notes.Chapter titles are Debroy's, not the text's — "the chapter headings we have given are subjective," with one qualification he supplies and that is worth keeping: "unless of course the colophon suggests a chapter heading." The risk is demonstrated, not merely stated: ch 434 is titled (Greatness of Panchakshara Mantra) and its body is dominantly about the six-syllable mantra — "six aksharas" ×7 against "panchakshara" ×2, opening on "the mantra with six aksharas" and closing on "'AUM Namah Shivaya.'" Note 1959 confirms the two are different mantras.
Roughly twenty-three disclosed editorial acts, and the handling is asymmetric and invisible. Fourteen located substitutions — including twenty-six → thirty-six Puranas ("but it should be 36"), two divine names replaced (Shambhu → Prabhu, note 2104; Hari → Hara, note 1925, which is doctrinally live), aksharas silently corrected in a mantra passage (note 2098, locations not given), and Vishvayogi → Vishvayoni because it "sounds more natural." Five unlocated "liberties," one covering an entire chapter (note 1881, ch 418). Four structural impositions — invented chapter titles, the thousand-name numbering added with repeats silently deleted, the 28 × 5 hell grid imposed on an undifferentiated list, and the 106 disciple-numbers added. Around seventeen divergences were fixed and around nine flagged and left standing, and the reader of the body cannot tell which was applied at any point.
A named editorial technique: demoting a proper name to an adjective to make a count come out. Note 1367 (rouchya) and note 1948 (Guhavasi). Both remove a numerical inconsistency; both are disclosed once.
The apparatus is a second voice on doctrine, not only on text. The corpus carries five mutually irreconcilable tattva counts — 24 (ch 424), 26 (ch 315, Kailasa L891), ~50 enumerated (ch 420) from the body; 31 (note 2086) and 36 (the Introduction) from the translator — and the Introduction's 36, which primes every reader, appears nowhere in 160 chapters of body.
The apparatus's errors are not only omissions. Note 1708 calls the thirty-eight Matrikas "puzzling" and measures the number against an external norm; the text derives 38 thirteen chapters earlier by partitioning Pranava's five components (8 + 13 + 8 + 4 + 5) and uses it eight times across both Samhitas.
⛔ The endnote layer is not uniformly commentary on this text. Note 1143 supplies Vimarshana's whole story — "a king of the Kirata kingdom… in the form of a dog, he died in front of a Shiva lingam" — which the Purana nowhere contains, in any volume, unmarked as external. Note 1857, asked for Sharada's story, points back to the litany that created the gap. Before citing any note as evidence, establish whether it is (a) commenting on this text, (b) importing from the tradition, (c) recording an emendation, or (d) hedged. 48 notes contain probably, presumably, seems or appears, including identifications this vault has cited as fact.
The volume narrates the same episode twice, repeatedly, and does not notice. The Sudarsana's origin (ch 305 / 307); Krishna and Upamanyu (ch 315–317 / 423, with three incompatible identities for Samba, one supplied by note 1918); the Kena yaksa with the deity swapped (ch 362 / 424); two Kausiki origins (ch 361 / 412); two etiologies for Sudyumna's sex, opposite in agent and valence (ch 308 / 350); two hell systems that disagree about whether hell is above or below the nether regions (ch 322 / 330); the headless shadow meaning death in a month and aphasia in six, in one unbroken conversation (ch 339 / 342); 26 tattvas against ~50 in the same rite with the same spoken formula (Kailasa L891 / ch 420); five Pranava components against four, stated twice on each side; the Prajapati roster with Shiva and Sati dropped between ch 404 and ch 452; and Anasuya "devoted to her husband" (ch 316) and "abandoned her husband" (ch 437). None of these is flagged anywhere.
Seven chapters and 373 shlokas in this edition's Sanskrit text are from the Skanda Purana, by Debroy's own statement, "since they concern the Shiva Purana." The eighth Samhita number — the 8.1 / 8.2 codes this vault has used across three volumes — exists only because of that graft, and he says so.
The disputed-Mahapurana tension carries forward and this volume sharpens it. Debroy quotes Hazra at length and calls his dissertation "still one of the best introductions to the Purana corpus" — and Hazra does not treat the Shiva Purana as a separate Purana at all, holding that "the attempt to raise the Siva-purana to the status of a Mahapurana was due to a comparatively late sectarian zeal," with the real Mahapurana being the Vayu (350–550 CE). Debroy prints the indictment first and at length, then argues back — "Who can determine which preceded which?" ⚑ This volume's largest section, 76 chapters, is the Vayaviya — named for the older text scholarship says it displaced. Any page treating this source as canonically equivalent to, say, the Vishnu Purana is choosing inside a live dispute and should say so. [SCHOLARLY — translator's introduction]
Three technical strata in this volume sit later than a 1000–1400 CE Purana is usually assumed to carry, and one of them is a citation rather than vocabulary. Khecari and kaki mudra taught by name and by method (ch 341); the full Pasupata / Shaiva Siddhanta apparatus — pasu, pasa, pati, the five kancukas, mala — in a recursive catechism (ch 392); the Shakta emanation chain Kriya Shakti → nada → bindu → Sadashiva (ch 426); Shaiva Siddhanta diksha in full physiology (ch 438); Vedic accent specification for the pancaksara (ch 434); the mātrā defined by a finger-snap (ch 460); the devayana/pitryana personified as ten deities and then worshipped as goddesses (ch 386–387); and the Shiva Sutras quoted by number and expounded, with note 1698 naming Vasugupta (ch 381). ⛔ This build does not adjudicate any of the resulting chronology questions. They are recorded as what is on the page and routed to the source map.
Two standing flags apply corpus-wide and are named on downstream pages rather than re-argued per page.
🚩 MOTIVATED REASONING — structural. This is a sectarian text whose interest is in itself and its god.
🚩 TRANSLATION DISTORTION — present, mitigated by disclosure, and the disclosure is the problem. Debroy's stated method is "(a) reducing interpretation; (b) relegating interpretation to footnotes and (c) when there are alternative interpretations, pointing this out to the reader through those footnotes." Clause (b) means the interpretive acts live in the apparatus by design — which is exactly where three volumes of decisive findings have come from. He also closes with "there is no substitute to reading these texts in the original Sanskrit."
⛔ Sectarian and violent material is reported as the text's own polemic, never as description of any practice, community or period. This governs ch 461 absolutely — its martyrdom doctrine, its licensed deaths by fasting and drowning, its abolished funeral rites, its permitted burial, and its redirection of a man's inheritance from his children to Shiva's devotees. It also governs ch 353's absorption of the Shakya line, which is reported as this text's handling and never as history.
⛔ Ch 338 is a near-verbatim Mahabharata import, announced by note 1299 — "the next chapter reproduces those sections from the Mahabharata, almost verbatim." It must never be cited as this text's own view of women.
Ch 359–364 carry the Devi Mahatmya's frame, its three episodes and their order, and no note of any kind attaches — while Debroy is himself the published translator of the Markandeya Purana. [PLAUSIBLE — needs corroboration]; the collation has not been run and no page may claim it has.
Naming caution. Guha denotes both the Nisada hunter renamed at ch 311 and Karttikeya (note 1918). Galava denotes two different figures (ch 317, ch 351). Note 1139 discloses that proper names are amended across volumes to harmonise them — so no finding in this vault may rest on a name-form alone.
Confidence tagging. Claims drawn from the body are claims about what this text says, cited by line reference and by Samhita-internal chapter code, and carry no web-source confidence tag. Claims about dating, transmission or scholarship drawn from the Introduction are tagged [SCHOLARLY — translator's introduction].
Evidence tagging for structural findings. Assertions this vault makes about the text carry [n=1 OBSERVED], [n=2 SUGGESTIVE] or [n≥3 PATTERN], and no rigour-register language below n=3.
Absence claims. All 160 chapters of this volume are covered and the endnote layer is fully censused; five saturation passes were run and the fifth returned zero new candidates. Where a finding rests on a scripted whole-file sweep rather than on line-reading, downstream pages say so.