Before Śiva gives Satī the nine limbs of bhakti, he gives her the classifications. There are a lot of them, they arrive in about four sentences, and they do not fit together.
*"O Devī! It is said that bhakti is of two types, saguṇa and nirguṇa, following the norms and natural. It is also said to be superior and inferior. Based on whether it is continuous or intermittent, there are also two differences. Of the two, it is said that continuous bhakti is of six types. The learned also know of differences based on whether it is prescribed, or not prescribed. Thus, there are many kinds of bhakti, described elsewhere. O beloved! As described by sages, know that there are nine aspects to bhakti."*1
Count them. Saguṇa/nirguṇa. Norm-following/natural. Superior/inferior. Continuous/intermittent — with continuous itself six-fold. Prescribed/not prescribed. And then nine aspects, on a completely different axis.
Six cuts and a nine-fold list in four sentences, and not one word about how any of them relate.
Try to build the grid and it collapses immediately.
If bhakti is saguṇa or nirguṇa, and superior or inferior, and continuous or intermittent, and prescribed or not — that is sixteen cells before you add anything. Add the six sub-types of continuous and you have more.
And none of the nine aspects sits anywhere in that grid. Hearing, chanting, remembering, tending, serving, worshipping, prostrating, friendly conduct, surrender — these are activities. The binaries are properties. A person doing śravaṇa could presumably be doing it continuously or intermittently, prescribed or not, in a saguṇa or nirguṇa mode. The nine cross-cut all of it and the text does not say so.
Worse, some cuts look like restatements of each other in different vocabulary. Superior/inferior against saguṇa/nirguṇa: is nirguṇa the superior one? The text does not say. Norm-following/natural against prescribed/not prescribed: those look like the same distinction twice, and they are given as separate cuts inside the same sentence-run.
The one place the text goes a level deeper is also the one place it stops.
"Of the two, it is said that continuous bhakti is of six types."
Six what? They are not named. Not one of them. It is the only sub-division in the passage that gets a number and no contents.
That is a specific kind of gap. A list without a number is vague; a number without a list is a citation — it points at something the speaker expects to exist elsewhere, in a form he is not reproducing.
And the very next sentence confirms it: *"Thus, there are many kinds of bhakti, described elsewhere."*
Two words doing an enormous amount of work.
Śiva, teaching the supreme tattva to Devī, refers her to another text.
The volume does this rarely. It is a text that prefers to contain things — it gives the eighteen Mahāpurāṇas, the pañca-lakṣaṇa, the thirteen saṃskāras, the eight forms, the five acts, the fifty-seven gaṇa-leaders. When it has a list, it lists.
Here it has at least six classificatory schemes, names none of the six sub-types, and defers.
The most economical reading is that this passage is a summary of something the compiler had and did not include — a doxographic paragraph, condensing several existing accounts of bhakti into a run of "it is said that" and "the learned also know of."
Note the attribution formulas. It is said. It is also said. The learned also know. As described by sages. Four separate hedges in four sentences. The nine aspects, by contrast, get defined one by one in the speaker's own voice.2
The build already holds the instrument for this, and this is its cleanest instance.
A tidy list tells you what its compiler believed. A lumpy one tells you what he inherited and would not discard. When a list will not sort, stop hunting its principle and count its groups.3
Count the groups here and you get, at minimum:
A Sāṅkhya-flavoured cut (saguṇa/nirguṇa). A ritual-law cut (prescribed/not prescribed; norm-following/natural). An evaluative cut (superior/inferior). A temporal cut (continuous/intermittent, six-fold). And an activity list (the nine).
Those are not one scheme with sub-branches. They are five vocabularies, from different discourses, laid end to end by someone unwilling to drop any of them.
And the tell is the attribution. The inherited material gets "it is said"; the nine get definitions. The compiler is confident about exactly one of the six schemes, and it is the one he explains.
The transferable technique is short and it works on any documentation you did not write.
When a classification will not compose, stop trying to compose it and read it as sediment.
Three moves.
Separate by vocabulary, not by content. Cuts drawn from different discourses will use different words for the same territory. Saguṇa/nirguṇa is metaphysics; prescribed/not prescribed is law; superior/inferior is evaluation. Three languages means at least three sources.
Find the one with contents. In any pile of schemes, one will be worked out and the rest will be named and dropped. The worked one is the author's; the named ones are inherited. Here: nine defined, six unnamed, four hedged.
Then check the hedges. It is said. The learned know. Described elsewhere. Attribution formulas cluster on borrowed material, and they are far more reliable than tone.
The payoff is not pedantic. A taxonomy you cannot compose is not a bad taxonomy — it is several good ones stacked, and the practical question changes from which cell am I in (unanswerable) to which scheme is this institution actually running (answerable, and usually the one with the definitions).
One consequence worth stating plainly, because it applies to this whole volume.
The build has repeatedly found lists in this text that will not resolve: the eight bonds of Prakṛti defined twice, differently, three hundred endnotes apart. The eight forms, of which the eighth is a role rather than a substance, with Debroy noting "there is thus no unique way to interpret the number eight." The sages made from body parts, whose taxonomy is "lumpy" enough that the build named it a stratigraphy.4
Four independent cases now, in four different chapters, of a count being load-bearing while its contents are negotiable.
That is a positive finding about how this text was made rather than a complaint about its rigour. The number is what was transmitted; the members were re-supplied by whoever was compiling. Which is exactly what you would expect of material that moved orally in a mnemonic frame and was written down more than once.
The last observation, and it is the reason the chapter still works.
Everything in the passage is inert except the nine.
The binaries cannot be acted on. Knowing that bhakti is saguṇa or nirguṇa, superior or inferior, prescribed or not tells a practitioner nothing about what to do on a given morning. They are descriptions of a thing already happening.
The nine are instructions. Each is defined in operational terms — sit steadily and drink in the account; visualise and recite; see him pervading everything; serve at sunrise; offer the sixteen upacāras; do sāṣṭāṅga praṇāma; believe that whatever happens is for your welfare; retain nothing for yourself.2
And they carry no counts, no ranking, and no cells.
So the passage runs a classificatory apparatus that cannot be entered and then hands over a practice that can be entered nine ways. The taxonomy is what the tradition had; the nine are what it could give somebody. The volume does not distinguish them, and the distinction is the whole value of the chapter.
Strongest support is the passage at l.1723: six classificatory cuts, a six-fold sub-division with no contents, "described elsewhere", and then the nine aspects on an unrelated axis — all inside four sentences.
Tension one, and it is the page. The cuts do not compose. Sixteen-plus cells before the six-fold, and the nine aspects belong to none of them — activities against properties, with no statement of how they relate.
Tension two. Two of the cuts look like the same distinction in different vocabulary — norm-following/natural and prescribed/not prescribed — given as separate divisions in one sentence-run.
Tension three. A number with no contents. "Continuous bhakti is of six types", and not one is named — followed immediately by "described elsewhere." A number without a list is a citation.
Tension four. The attribution formulas sort the material. "It is said," "it is also said," "the learned also know," "as described by sages" — four hedges on the inherited schemes; the nine get definitions in the speaker's own voice.
Open question. Is the elsewhere anywhere? The volume claims five deleted Saṃhitās and admits its own transmission is lossy.5 If no chapter in this Purāṇa names the six types of continuous bhakti, then "described elsewhere" points outside the surviving text, which is a small but hard datum about what the compiler had in front of him. A single targeted search would settle it.
The Nine Limbs of Bhakti is the same passage's other half. The taxonomy is what the tradition had; the nine are what it could hand somebody — and the volume does not mark the difference.
The Eight Bonds of Prakṛti carries two incompatible definitions from two endnotes three hundred apart. Second case of a count surviving while its contents are re-supplied.
The Aṣṭamūrti: Eight Forms has an eighth member that is a role rather than a substance, with Debroy conceding "there is thus no unique way to interpret the number eight." Third case.
The Sages Made From Body Parts is where the build first named the instrument — an inconsistent taxonomy is a stratigraphy. This passage is its cleanest specimen: five vocabularies, one worked scheme, four hedges.
The Blemishes Taxonomy: Failure Modes (creative-practice). The Nāṭyaśāstra enumerates what can go wrong in performance — the crown slipping, a mis-struck drum, a stray dog on stage, an actor forgetting a line, cow-dung thrown by a rival company, a meteor. A named, ordered, teachable doṣa taxonomy, produced by roughly the same culture in roughly the same period as this Purāṇa.
Set them side by side and the difference is not sophistication. It is who the list had to reach.
The Nāṭyaśāstra's taxonomy sorts because it is a training document for a guild — it has to be transmitted to strangers who will apply it without the author present. Every category has to be distinguishable by someone who was not there.
The bhakti taxonomy does not sort because it never had to. It is a doxographic summary, delivered inside a narrative frame, to an interlocutor who already knows the answer.6
What the pair produces: a taxonomy's internal consistency is a measure of its transmission requirements, not of its author's rigour — which the build recorded once before and can now state with two cases from one culture.3 Which yields a usable inversion: where you find a lumpy classification, do not conclude the field was confused. Ask who was going to have to use it — and expect the answer to be nobody outside the room.
Authorship and Dating Debate (history). You date an undated text by what it mentions. The Arthaśāstra names china-paṭṭa, so it postdates 221 BCE; red coral from Alexandria, so it postdates the monsoon trade; Parasamudra for Sri Lanka, attested around 100 CE; silver and copper coins but no gold, so before the Guptas. Pile up the internal evidence and a date emerges. Maybe.
Chapter 75's classifications are the same evidence type, in a domain where nobody usually looks for it.
The trade-goods method dates by objects mentioned. This dates by schemes deployed. A passage carrying a Sāṅkhya cut, a ritual-law cut, an evaluative cut, a temporal cut and an activity list is carrying five discourses — and discourses, like trade goods, have periods.
The insight neither gives alone: classification vocabulary is a dateable artefact, and it is more robust than object-mentions because it is harder to interpolate. A copyist can add a place-name in one word. Adding a whole classificatory scheme, with its own vocabulary and its own hedging formula, is a much larger intervention — and it leaves the seam visible, as here, where five schemes sit end to end refusing to compose. Which suggests a method this build has been approaching from the other side: the layers of a composite text may be readable off its taxonomies more reliably than off its stories, because a story can be retold smooth and a borrowed classification arrives with its vocabulary attached.
Sharpest implication. In four sentences before the nine limbs, Śiva gives Satī six classificatory cuts of bhakti — saguṇa/nirguṇa, norm-following/natural, superior/inferior, continuous/intermittent, continuous being six-fold, and prescribed/not prescribed — and not one word about how any of them relate. Build the grid and it is sixteen-plus cells before the six-fold, and the nine aspects belong to none of them: activities against properties, cross-cutting everything. Two of the cuts look like the same distinction in different vocabulary. And the six-fold sub-division has a number and no contents — "continuous bhakti is of six types", none named — followed immediately by "described elsewhere." A number without a list is a citation. Read the attributions and the passage sorts itself: "it is said," "it is also said," "the learned also know," "as described by sages" — four hedges on the inherited schemes, while the nine get defined one by one in the speaker's own voice. Five vocabularies from five discourses, laid end to end by someone unwilling to drop any of them, with exactly one worked out. The taxonomy is what the tradition had; the nine are what it could hand somebody — and the volume does not distinguish them.
Generative questions.