A scripted census of this volume's technical vocabulary returns a number worth sitting with.
116 distinct technical terms carry an endnote gloss in the prose. 111 of them are glossed exactly once.1
Once. The term appears, the translator explains it, and neither the term nor the explanation recurs anywhere in a hundred and sixty chapters.
⚑ Most of the 111 are what you would expect — a measure, a flower, a metrical name. But not all of them. Some carry real doctrinal weight: a named eight-fold subtle body that is the vehicle of transmigration in one Śaiva system; a named offence about marrying out of birth order; a named technique for joining two people's nāḍīs; a named foundation-stone; a named concluding bath.
Each appears once, is glossed once, and does nothing else.
And a disproportionate number of them sit in chapters an earlier pass of this build never opened.
The instrument, stated plainly: every technical term in the body carrying an endnote gloss was extracted by script and counted by number of occurrences.1
⚑ A ninety-six per cent hapax rate in the glossed vocabulary. That is a fact about the edition's annotation practice as much as about the text — Debroy glosses on first appearance, and if a term never reappears the gloss never reappears either.
⛔ So the census measures two things at once and cannot separate them: how often the text uses a term, and how often the translator felt one needed explaining. What it does isolate is the intersection — terms that occur once and were judged to need a note. That set is where the volume's unabsorbed technical material lives.
The ones that matter, each located and each read in place:2
| Term | Where | Note | What it is, on the page |
|---|---|---|---|
| puryaṣṭaka | ch 419-8.1(31) | 1895 | the eight-fold subtle body, which "rises up through sushumna" at death |
| parivetta / parivitti | ch 320-6(6) | 1182 | marrying, or being married, out of birth order — a named pair of roles, listed among the sins |
| brahmakūrca | ch 433-8.2(11) | 1956 | a purificatory drink, prescribed "every month" |
| brahmaśilā | ch 458-8.2(36) | 2090 | the foundation-stone laid in the pit at a liṅgam installation |
| nāḍīsandhāna | ch 438-8.2(16) | 1986 | the guru's technique at initiation — "the union of the nāḍīs" |
| avabhṛtha | ch 462-8.2(40) | 2100 | the concluding bath of the frame-sacrifice, in the volume's penultimate chapter |
Plus the mantra-apparatus vocabulary, each named once: śaḍutthāpana (ch 387, note 1770), and kīlaka, bīja-śakti and digbandha together in one list at ch 433 (note 1963), and saṃpuṭa at ch 442 (note 2000).2
⚑ These are not ornaments. Parivetta is an offence with a name and a counterpart. Nāḍīsandhāna is the operation by which one person's consciousness is transferred into another. Puryaṣṭaka is what survives a death. Each is the sort of term a tradition builds a section around, and each occurs once.
The clearest case, because the term is load-bearing everywhere else it exists and inert here.
Ch 419-8.1(31) describes what happens when the knots of the kalās are severed by Śiva's names:
"When they are severed in this way, puryashtaka rises up through sushumna, up through the heart, the throat, the palate, the middle of the eyebrows and Brahmarandhra. It will be established in Shiva's energy, above the twelve different joints and the moon there."3
⚑ A complete ascent, with six stations named, of a thing the volume never otherwise mentions.
⛔ And the volume nowhere says what the eight components are. In the Śaiva Siddhānta systems the term belongs to, the puryaṣṭaka is the subtle body that transmigrates — the vehicle that carries karma between births. This volume uses the word, routes it up the suṣumnā, and does not enumerate it.
⚑⚑ Which is the more striking because the volume enumerates everything else. It gives ten prāṇas by name, five kañcukas, thirty-eight mātrās derived from a five-part partition, sixty-four siddhis across eight tiers with every carry-forward stated.4 A text this committed to enumeration, using an eight-fold technical term and not listing the eight, is a text that has inherited the term rather than built it.
⚑ And it lands in a live gap: the volume's account of what survives death is otherwise supplied by the birth-canal passage — memory destroyed by the delivery — and by the fourteen-species ritual exhaustion at ch 440. Puryaṣṭaka would be the mechanism joining them, and it is never brought to either.5
The distribution is the method finding, and it is uncomfortable.
Four of the six weighted terms are in chapters the build's first pass never line-read — 419, 433, 458, 462 — and were recovered only by a later sweep.1
⚑ The reason is structural. An instrument tuned to doctrine, narrative or contradiction finds nothing in a ritual-procedure chapter, because a procedure chapter contains no propositions to catch on. "After it has been cleaned, it will be smeared with sandalwood paste and placed in the pit" trips no detector.
⚑⚑ And a once-glossed term is invisible to a reading pass as well, for the opposite reason: a reader meets an unfamiliar word, clicks the note, understands the sentence, and moves on. The term is absorbed rather than noticed. It takes a count to see that it never came back.
⛔ This is the same failure mode that cost this build the four cows, the Śākya entry and Sanatkumāra's camel — all in ranges the sweeps scored zero on. Instruments find what they are shaped for. The once-glossed layer is shaped for nothing.
One of the six deserves its own paragraph, because of where it falls.
The avabhṛtha is the concluding bath of a sacrifice, and it appears in the volume's penultimate chapter, as the frame-sacrifice at Naimiṣa ends:
"The sacrifice in Naimisha was over. When it was morning, all the stores of austerities got ready to perform the avabhritha rites. The delighted Devi Sarasvati followed the instructions of Brahma himself. She flowed in the form of an auspicious river that was full of tasty water. Seeing River Sarasvati, the minds of the sages were pleased."6
⚑ A goddess is instructed to become a river so that the audience of the Purāṇa can have somewhere to bathe at the end of it. The term for the rite is glossed once, in note 2100, and never occurs again.
⚑⚑ And it closes the volume's frame in a way the vault should record: the entire Purāṇa has been narrated inside a sacrifice, and the sacrifice has a concluding rite, and the concluding rite happens. ⛔ Which is a small point and a real one — the frame is not abandoned. It is discharged.
Stated carefully, because the temptation to over-read is strong and the evidence is thin in a specific way.
A term used once can mean several things. That the text inherited a vocabulary it does not systematically deploy. That the abridgement cut the sections where the term was at home. That the compiler knew the word and not the system. Or simply that one occasion required it.
⚑ What distinguishes the six above from the other 105 is that each has a system behind it elsewhere — an eightfold subtle body, a pair of marriage-order offences, a consciousness-transfer technique. A term with a system behind it, used once and not enumerated, is the strongest available sign of inherited rather than generated vocabulary.
⛔ [n=6 OBSERVED] on the weighted subset. No claim is made about which of the four explanations applies to any of them, and the volume itself supplies a candidate for one: it documents the excision of five whole Saṃhitās, fifty-two thousand shlokas, by name.7 ⚑ A term whose home section was cut would look exactly like this — and so would three other things.
You have a corpus and you want to find what it inherited rather than built.
Count the hapaxes in the glossed vocabulary, not the whole vocabulary. ⚑ Every corpus has thousands of words used once. The interesting set is the intersection of used once and judged to need explaining — because the second condition is somebody's estimate of what a competent reader would not know, and a term meeting both is one the corpus is using without owning.
Then check which of them have a system behind them elsewhere. A once-used measure is a once-used measure. A once-used term that names an eight-part structure, in a corpus that enumerates everything, and whose eight parts are never given — that is a borrowing. The tell is the missing enumeration, not the frequency.
Look at where they cluster. ⚑ Here four of six are in procedure chapters, which are exactly the chapters that defeat every detector — no propositions to catch on, and a reader who clicks the note and moves on. It takes a count, not a read, to see that a word never came back.
And resist the single explanation. Inherited vocabulary, an excised source section, a compiler who knew the word and not the system, or one occasion that needed it — all four produce the same signature. The useful output is the located list; the story is not available from the count.
Evidence. Scripted census of glossed technical terms across the body: 116 distinct, 111 hapax.1 All six weighted terms located to chapter and endnote, with their surrounding sentences read in place.236 The mantra-apparatus vocabulary located likewise.2 The distribution against the build's own coverage record checked chapter by chapter.1
Tension 1 — the census measures the annotation as much as the text. ⛔ Marked in the body. Debroy glosses on first appearance, so a hapax gloss follows automatically from a hapax term, and the count cannot separate the translator's judgement from the text's usage. ⚑ What survives is the located set, which is what the page actually uses.
Tension 2 — "a system behind it elsewhere" is knowledge this build brought, not evidence it found. ⚑ That puryaṣṭaka is load-bearing in Śaiva Siddhānta is a fact about that tradition, and this build has not verified it against a source. ⛔ The claim on the page is bounded to what is observable here: the volume uses an explicitly eight-fold term and never gives the eight, which is a fact about this text alone.
Tension 3 — five terms recur and the page says nothing about them. ⚑ The census returns 116 minus 111 = five recurring glossed terms, and this build has not examined which. They would be the natural control — terms the volume did absorb — and the comparison has not been run.
Open question. The volume documents the excision of five Saṃhitās totalling fifty-two thousand shlokas, by name. ⚑ Do any of the once-glossed weighted terms belong to the subject-matter of an excised section? Puryaṣṭaka and nāḍīsandhāna are initiation-and-transmigration vocabulary; a Rudraikādaśa or a Dharmasaṃjña at thirteen and twelve thousand shlokas would have had room for both. ⛔ Unfalsifiable on present evidence — the sections are gone — and worth stating as the shape a real answer would have.
Against The Siddhi Cascade: the contrast is the argument. That page's passage enumerates sixty-four items across eight tiers with every carry-forward stated. ⚑ This page's puryaṣṭaka is eight-fold by name and never enumerated at all — in the same Saṃhitā, forty-one chapters apart. The volume's enumerating habit is strong enough that a failure to enumerate is itself a signal.
Against The Five Lost Saṃhitās: the natural home for the excision hypothesis, and the reason it cannot be tested. ⚑ A corpus that names what it removed gives you a candidate explanation for its orphan vocabulary and no way to check it.
Against The Apparatus Is the Translator's: this page depends on the apparatus more completely than any other in the build. The set exists because Debroy decided which terms needed notes. ⚑ A different translator would produce a different 116, and the page says so rather than treating the census as a property of the text.
Against The Guru's Cow: the same method finding at a different scale. Four cows in six unread lineage chapters; four weighted terms in four unread procedure chapters. ⚑ Both were invisible to instruments tuned to doctrine, and for the same reason: neither a genealogy nor a ritual procedure contains propositions.
AI Second Brain Retrieval Shift — that page's claim is that meaning-based recall dissolves the old filing problem: you no longer have to guess your past self's taxonomy, because retrieval runs on what a thing means rather than on where it was put. It names the collector's fallacy — hoarding notes nobody ever looks at again — as the pathology the old architecture produced.
The once-glossed layer is that problem at its purest, and it supplies a failure mode the retrieval frame usually treats as solved.
Nothing here is lost. Every one of the 111 terms is present, in a searchable file, with an explanatory note attached, correctly located. A reader who wants puryaṣṭaka can find it in a second. ⚑ The failure is that nobody knows to want it — because the term was absorbed on first reading, the note answered the local question, and the fact that it never recurred is a property of the whole corpus that no single encounter reveals.
⚑⚑ That is a retrieval failure with no missing data and no bad index. It is a failure of query formation: the thing you would need to search for is a frequency, and frequencies are not what anybody searches. Which is why the census was the instrument and reading was not — and why four of the six were in chapters the build had read past twice.
The transferable diagnostic follows, and it is uncomfortable for exactly the architecture that page celebrates: meaning-based recall does not touch this failure, because the query is never formed. ⚑ The collector's fallacy names material captured and never revisited; this is material captured, revisited, understood, and still never seen — because what would have to be noticed is a frequency, and no semantic search returns a frequency. The only instrument that finds it is a count over the whole store, and counts are the one operation a well-retrieved store makes nobody run, because everything is already findable.
Sharpest implication. A scripted census of this volume's glossed technical vocabulary returns 116 distinct terms, of which 111 appear exactly once. Most are measures and flowers. Six are not. Puryaṣṭaka — the eight-fold subtle body — is routed up the suṣumnā through six named stations at ch 419, ⚑ and the volume never says what the eight are, in a text that gives ten prāṇas by name, five kañcukas, thirty-eight mātrās from a five-part partition and sixty-four siddhis across eight tiers with every carry-forward written out. Parivetta and parivitti name a paired offence about marrying out of birth order, once, in a sin-list. Nāḍīsandhāna names the operation by which a guru transfers consciousness into a disciple, once. Brahmakūrca, brahmaśilā and avabhṛtha likewise — the last of them in the penultimate chapter, where a goddess is instructed to become a river so the audience can bathe at the end of the Purāṇa they have been listening to. ⚑⚑ Four of the six are in chapters an earlier pass never opened, and a once-glossed term is invisible to reading as well as to sweeps — the reader clicks the note, understands the sentence, and moves on. It takes a count to see that a word never came back.
Generative questions.