The two seats: how the tantric canon was actually organised

On a Sunday in January of 1052, in the quarter of Kathmandu that surrounds the temple of Pashupatinath, a scribe named Jayākarajīva finished copying a book onto palm leaves. Three hundred and fifty-eight folios. He was copying a tantra called the Brahmayāmala, also known as the Picumata.

Every other surviving copy appears to descend from his. No complete manuscript appears to survive outside Nepal.

It is a strange object to carry the weight it carries. Much of what we can say about how the tantric scriptures classified themselves at the earliest recoverable stage runs through that one codex, catalogued today as NAK 3-370 and microfilmed by the Nepal-German Manuscript Preservation Project as A42/6. And on the specific question this article is about, that codex disagrees with the figure printed in the standard reference works.

The figure everyone prints

Open any survey of tantra and you will meet the same diagram. Śaiva revelation divides into streams. One of those, the Bhairava stream, divides into two seats: the Mantrapīṭha, the seat of the masculine mantra-deities, which the handbooks gloss as headed by the Svacchandatantra; and the Vidyāpīṭha, the seat of the feminine vidyā-goddesses, the stratum from which much of later Śākta religion descends. The Vidyāpīṭha then branches into the Yāmala texts and the Śakti-tantras, and the Śakti-tantras branch again into the tantras of Kālī and the Trika, or, in Sanderson’s own careful parenthesis, “of what was later called the Trika” (1988, p. 669).

It is a good diagram. It is genuinely useful. Students learn it, and they should.

Which Sanskrit text draws it?

None of them draws the whole of it. The two-seat division at the top is stated in a primary source, and we will come to where. The branching beneath it is a modern reconstruction, assembled across scattered and mutually disagreeing texts, and the scholar who assembled it said as much in print, in the same discussion that carries the figure.

Two seats, or four

Start with the count, because the count is where the tidy picture first cracks. The word pīṭha means a seat, and the tantras that use it as a category of scripture usually name four.

The Brahmayāmala defines them in its thirty-eighth chapter, in verses Shaman Hatley edited in the notes to his 2007 dissertation: those scriptures that rest on vidyās are the Vidyāpīṭha, those that rest on mantras the Mantrapīṭha, those that rest on mudrās, ritual gestures, the Mudrāpīṭha, those on maṇḍalas the Maṇḍalapīṭha. Four seats. Jayaratha, commenting on the Tantrāloka in the thirteenth century, writes that scripture is fourfold by seat: catuṣpīṭhaṃ tāvac chāstram.

Even the passage that gives us the two-seat formula gives four. Abhinavagupta writes in the thirty-seventh chapter of his Tantrāloka that the corpus is divided in two, right and left, Mantra and Vidyā. Then, in the next pāda, he derives the other two from them: mudrāmaṇḍalagaṃ dvayam, the pair belonging to Mudrā and Maṇḍala. He is not offering two seats as a rival to four. He is giving the genealogy of the four.

The two-branch figure keeps the first half of that sentence and drops the second.

The four are not four equals, either. The last two are thinly populated. Mark Dyczkowski, whose Canon of the Śaivāgama of 1988 is the one book-length account to set beside Sanderson’s, found that the Brahmayāmala says simply that the Mudrā and Maṇḍala seats contain all the gestures and diagrams of all the tantras in all the streams. The Jayadrathayāmala refers to a single, unnamed root text for the Maṇḍalapīṭha and then says it is present in the other three anyway. In the Jayadrathayāmala the Mudrāpīṭha does slightly better, with three root tantras, one of which is the Kubjikāmata, a major scripture in its own right.

A formula reported from a book that no longer exists

The clean statement of the two-seat division occurs at Tantrāloka 37.18c–19c1 in Sanderson’s numbering, which corresponds to 37.18–19a in the GRETIL transcript of the Kashmir Series text. Abhinavagupta is reporting there, not asserting. The line runs śrīmadānandaśāstrādau proktaṃ bhagavatā kila: it has been taught by the Lord in such scriptures as the Ānanda. Sanderson’s judgement is that Abhinavagupta is “quoting or paraphrasing” a lost Ānandaśāstra (2009, p. 45 n. 11), and the citation formula itself, whose ādau means “beginning with,” points to that text as first among several rather than as the sole authority. He invokes the same source earlier in the same chapter for a different point.

The Ānandaśāstra is lost. There is no manuscript of it.

That is worth pausing on, because it is how a great deal of what we know about the Āgamas and Tantras survives at all, and it is the same question the tradition works through in Āgama and Veda: what counts as revelation, and on whose authority.

The book that never claimed its throne

The Brahmayāmala is older than Abhinavagupta by two centuries or more. Sanderson grants that the date “is still a matter of debate,” then judges it “unlikely to have been composed later than the end of the seventh century or earlier than the sixth” (2009, p. 51). Hatley is more open: he finds it “highly likely” that the text, “in a form close to that preserved in its oldest Nepalese codex,” existed at some point in the eighth century, sees “no grounds for ruling out a seventh-century dating, especially for the older portions,” and holds that a sixth-century form “merits consideration” (2007, p. 228). He adds that it is “not impossible that the BraYā continued to develop into the early ninth century,” a possibility he ties to the dating of the Laghuśaṃvara and Tantrasadbhāva. The disagreement is real and neither party has closed it.

What matters here is what that early text does with the Svacchanda.

It puts it in the Vidyāpīṭha.

Not as an afterthought, and not by inference. In chapter thirty-eight, immediately after the verses defining the four seats, the text lists eight tantras named for forms of Bhairava, and the first of them is svacchandabhairavaṃ devi. Hatley prints the Sanskrit and his table of the Bhairava stream puts the Svacchanda at the head of the Vidyāpīṭha column. Dyczkowski, working independently nineteen years earlier, produced the identical list of eight, item for item.

Hatley draws the consequence (2007, p. 148):

“In some respects, the Vidyāpīṭha/Mantrapīṭha division appears contrived: note that the BraYā places the Svacchandatantra at the head of the Vidyāpīṭha as the first of eight tantras named after particular forms of Bhairava. Only four scriptures are assigned to the Mantrapīṭha, none of which appear extant.”

Both halves of that passage carry weight, and the second is the one usually lost. Dyczkowski, working from the same passage in 1988, counted seven, writing that the Brahmayāmala “appears to be uncertain of the contents of this pīṭha” (1988, p. 53) and marking two entries incomplete. Hatley counts four. Neither engages the other on this count, and this article cannot settle it. What both readings agree on is that the Svacchanda is not among them.

Honesty requires the other side of Hatley’s paragraph too. He introduces it on the same page by observing that the fourfold classification “appears to mask what Sanderson identifies as a more fundamental twofold division between the Mantrapīṭha and Vidyāpīṭha” (p. 148). The witness this section leans on is, in the same breath, a witness for the scheme it complicates. The picture is genuinely two-sided.

So where does the Svacchanda’s Mantrapīṭha membership come from? Not from the text itself, and not from a commentator’s invention. It is scriptural. The Jayadrathayāmala names it first of the Mantrapīṭha’s four in the thirty-sixth chapter of its first ṣaṭka, in a verse that appears to quote the now-lost Sarvavīratantra, and Kshemaraja cites the same verse from the same lost source in his commentary on the Svacchanda. The chapter colophons of the Nepalese manuscripts of the Svacchanda carry the assignment as well. What the Svacchanda declines to do is claim any single seat for itself as a whole, though at 14.26ab, closing a run of material on ritual gesture, it calls that material the Mudrāpīṭha.

A caution before this goes further, because two similar-sounding claims fuse easily and are not the same claim. The Svacchanda does say that it contains all four seats. That statement belongs to the Goddess’s opening question, describing the vast original text she is asking to have condensed, and Kshemaraja later reads it as a remark about the topics the surviving text covers. The text that contains all four seats in the Brahmayāmala’s own chapter is the catuṣpīṭhaṃ bhairavam, which Hatley notes is identified with the Brahmayāmala elsewhere but never in this chapter, where the Brahmayāmala appears only as the third of eight Yāmalas in the Vidyāpīṭha (2007, pp. 266, 268). To say the early tradition treated the Svacchanda as transcending the classification would move a claim from one book to another.

Two seats, two social worlds

None of this was a librarian’s problem. The seats were two ways of living.

The difference is one of degree, and the degree is transgression. The Bhairava Tantras carry a Kāpālika background, and Sanderson describes their basic ascetic form as “that of the skull (kapālavrata/mahāvrata)” (1988, p. 670). In the cult of Svacchandabhairava that background is present and muted: the god is, in Sanderson’s words, “transitional in type between the calm Sadāśiva of the Śaiva Siddhānta and the gods of the Kāpālika mainstream” (1988, p. 670). In the Vidyāpīṭha it is the reverse: where for Svacchandabhairava the cremation-ground culture was “somewhat in the background,” Sanderson writes, “here it is pervasive” (p. 670), and access to the powers of its deities ran through adopting the observance of the skull-bearers. The married householder is largely absent from the earliest scriptural sources, or subordinate in them.

Then the paradox. The majority Śaiva cult of medieval Kashmir was not that of the Śaiva Siddhānta. It was the cult of Svacchandabhairava, drawing its authority from a Mantrapīṭha scripture and therefore, in Sanderson’s phrase, “strictly speaking, a tradition of the Kāpālika-based left” (1988, p. 700). That is a third sense of “left,” meaning the non-Saiddhāntika traditions at large, and it is worth keeping distinct from the other two: Sanderson warns in a footnote that the left and right of the two seats “must not be confused with” the right and left currents (2009, p. 46 n. 12). Sanderson’s summary is dry: “the Kashmiri practised a thoroughly domesticated form of the cult” (p. 700). The agent of that domestication was Kshemaraja, working by commentary, and Sanderson does not think the result was faithful: “In purely doctrinal terms Kṣemarāja’s commentaries do violence to both of these texts… Neither Tantra fits either exegetical straitjacket” (p. 700).

Which is the point this section exists to make. If the line between the seats is a line of degree rather than kind, then it is exactly the sort of line a text can be shelved on either side of, depending on who is doing the shelving and when. That is not a flaw in the sources. It is what a gradient looks like when later readers need a boundary.

A claim about rank

Abhinavagupta reports the two-seat formula in the course of ranking what it divides. He puts Maṇḍala lowest, then Mudrā, then Mantra, then Vidyā highest. He makes the Siddhayogeśvarīmata the chief text of the Vidyāpīṭha and the Mālinīvijayottara the essence of that. And then, on Dyczkowski’s reading, he uses the authority of the lost Ānandaśāstra to make the Vidyāpīṭha the source of the rest. Dyczkowski translates the verse itself exactly as this article did, with the pair yielding the other pair (1988, p. 52); the stronger claim rests on the lines that follow, where the vidyā strengthens and nourishes the mantra.

How far that goes is worth stating carefully, because it is easy to overstate. At 37.21 the four seats are anyonyasaṃbhedavṛtti, working by mutual interpenetration, each containing all the others with one predominating. That is a standard move and not an abolition, and Abhinavagupta goes straight on using the scheme two verses later. Jayaratha, glossing him two centuries afterwards, gives the strongest version of it: in Dyczkowski’s rendering, “there is only one pīṭha which is of the nature of them all” (1988, p. 55). The same Jayaratha still writes that scripture is fourfold by seat. The scheme is being ranked and subordinated, not dismantled.

What the hierarchy was for is visible in how freely it was rebuilt. Dyczkowski put the principle plainly: “a text that records the structure of the canon is free to interfere with it” (p. 15). Hatley makes it structural, not accidental, arguing that this model of revelation “dictates that a new scripture claim for itself a privileged position on a hierarchical scale of texts” (2007, p. 268). Sanderson built the same point into his figure, noting that whatever stands above and to the left regards whatever stands below and to the right as lower revelation.

The scheme is a claim about who outranks whom, written by parties to the dispute.

What the manuscripts decided

So why did one seat come down to us as a single famous book, and the other as a library almost nobody has read?

Not doctrine. Survival.

The Mantrapīṭha, in Hatley’s description, is “a scriptural category otherwise poorly represented in the surviving literature, having this text alone as its major early exemplar” (2007, p. 148). The text he means is the Svacchandatantra. One book carried the category, and that book did well. In Sanderson’s description it “survives in Kashmirian, Nepalese, and South Indian manuscripts” and “attracted learned exegesis” (2007, p. 234), and in the early twentieth century it was printed with Kshemaraja’s commentary in the Kashmir Series of Texts and Studies.

The Vidyāpīṭha went the other way, and it had more to lose. Dyczkowski observed that most of the tantras preserved in Nepal which affiliate themselves to a seat belong to this one. The Jayadrathayāmala, whose seat-classification is the fullest that survives, has still not been critically edited in full. Chapter thirty-eight of the Brahmayāmala, the canon chapter, has never been published: the volumes that have appeared cover other chapters. The digest that preserves the lost Śrīkaṇṭhīyasaṃhitā’s account of the canon sat in a paper manuscript catalogued under the wrong title.

The asymmetry is not that one seat was printed and the other was not. Kashmirian Śaiva printing went both ways: the Tantrāloka with Jayaratha’s commentary ran to twelve volumes in the same series, and the Mālinīvijayottara, which Abhinavagupta’s own ranking puts on the Vidyāpīṭha side, was printed there in 1922. The asymmetry is that the Mantrapīṭha had one surviving book and it was read continuously, while the Vidyāpīṭha had a library and most of it is still in manuscript.

One debt should be recorded before leaving this. The Kashmir Series editions were the work of Kashmiri pandits, Madhusudan Kaul Shastri chief among them, and without their edition of the Tantrāloka the citation of the lost Ānandaśāstra would not be in general circulation today. Dyczkowski’s own acknowledgements name Navjivan Rastogi of Lucknow University, whose Krama Tantricism of Kashmir of 1979 he cites as an authority, and Vrajavallabha Dvivedi of Sampurnananda Sanskrit University; he closes the first part of his book quoting K. C. Pandey on how Śaivāgamic study should be done at all.

The scholars read what the editors saved.

The honest shape of it

Some things here are firm. The seat-classification is real, native to the texts, and used by them: the Brahmayāmala and the Jayadrathayāmala both place themselves in the Vidyāpīṭha in their colophons. The distinction between masculine mantra-deities and feminine vidyā-goddesses is genuine. The Siddhāntāgamas are excluded from the scheme entirely.

Other things need labelling as inference. In the Brahmayāmala the four seats sit inside a single stream, the right. It is later, once the Jayadrathayāmala extends the Vidyāpīṭha across the left as well, that the two schemes begin to cross-cut instead of nesting: only part of the Vidyāpīṭha is then assigned to the left, and the first ṣaṭka of the Jayadrathayāmala describes itself as ubhayātmakam, of the nature of both. Anyone drawing one clean nested tree across the whole corpus is drawing something no single text sustains. And the surviving account of the Śrīkaṇṭhīyasaṃhitā, which lists the sixty-four Bhairava Tantras in eight groups of eight, appears to carry no seat-division at all, though it survives only as an excerpt and absence from an excerpt is not absence from the whole.

Sanderson has been fair about this from the beginning and deserves the courtesy of being read closely. Introducing the classification in 1988, he wrote that these tantras “have been variously listed and classified in different parts of the corpus,” and that the classification he gives “corresponds, I believe, to the main structure of the Śaiva tradition outside the Śaiva Siddhānta at the time when the Kashmiri began their work of postscriptural systematisation in the ninth and tenth centuries” (p. 668). In the prose beside the figure, though not in the figure itself, he marked one branch of his own scheme artificial: the Tumburu division, “fitted into this scheme as a third division of the Vidyāpīṭha. But this is artificial” (p. 669). When he restated the scheme in The Śaiva Age in 2009 he kept that division and dropped a different node, the Yāmala and Śakti-tantra branching beneath the Vidyāpīṭha, which became a flat list of four sub-corpora.

The scheme was never wrong. It was scoped by its author to a particular moment, the point at which the Kashmirian exegetes began their work, and then reproduced for nearly forty years by people who kept the picture and dropped the sentence standing beside it.

There is a detail from the end of the tradition that says the rest. Jayaratha, in thirteenth-century Kashmir, is the greatest commentator the Tantrāloka ever had. He glossed Abhinavagupta line by line, two centuries after him, including his citations of the Brahmayāmala.

Sanderson reports that Jayaratha had no access to that text (2007, p. 237 n. 23).

The greatest commentator the tradition produced was, at the end of it, glossing citations of a book he could not open.

The classification is not the tradition. It never was.