The Formula Count

An Impression, Turned Into a Number

Anyone who reads a long stretch of this text develops the same impression: the phrases repeat. Somebody prostrates himself, somebody is an ocean of compassion, somebody is affectionate towards his devotees, and there is no doubt about it.

Impressions of that kind are cheap and unreliable — you notice repetition partly because you are looking for it, and a reader who was not looking would report a different feel. So the impression was converted into counts, run over the volume's narrative body: chapters 171 to 304, 165,512 words.

The numbers came back, and they are not subtle.

The Seven, and What They Are Counted Against

All counts run over body lines only, endnotes excluded, and re-run independently for this page rather than carried from the extraction record.¹

Second-tier constructions from the same sweep: "O lord among sages" 44 · "joined his hands in salutation" 49 · "supreme among sages" 37 · "Hearing these words" 25 · "showered down" 23 · "was pleased in his mind" 20 · "like a rod" 11.¹

⚑ The denominator matters as much as the numerator. 99 instances in 165,512 words is one every 1,670 words — roughly once every chapter and a half, sustained across 134 chapters. That is not a phrase somebody liked. That is a slot.

They Are Spread, Not Clumped

This is the check that turns the counts into evidence, and it is the one most people skip.

A high count can mean one of two very different things. It can mean one passage repeats a phrase heavily — a litany, a stotra, a list — which would tell you about that passage and nothing about the register. Or it can mean the phrase is distributed across the work, which tells you about the composition.

The distribution says the second, clearly. "Prostrated himself" appears in 55 of the 134 chapters in range — 41% of them. "Affectionate towards his devotees" in 42. "Sought refuge" in 42. "There is no doubt" in 41. "Ask for a boon" appears 25 times across 23 chapters, which is almost exactly one per chapter that uses it at all.¹

⚑ These are not concentrated in the hymns, where you would expect stock phrasing and where it would prove little. They run through narrative, dialogue and frame alike, in roughly a third to a half of all chapters. The formula is the register, not a feature of one genre inside it.

Analytical Case Study — Two Slots, Doing Two Different Jobs

The seven are not all the same kind of thing, and separating them sharpens what is being measured.

"There is no doubt" is a closer. Sixty-seven instances, 41 chapters. It carries no information — it does not add evidence, name a source, or qualify anything. It terminates a claim, and it appears wherever a speaker needs a sentence to stop with emphasis. Ch 176 alone stacks it: "This is the truth. There is no doubt that this is the truth."²

"Affectionate towards his devotees" is an epithet slot. Sixty-three instances, 42 chapters. Where a Homeric line needs a metrically-correct adjective for Achilles, this text needs a relational one for Śiva, and it has a default. The slot is filled almost every time the god is described as acting rather than speaking.

⚑ Those are the two classic signatures of a composed-by-formula register: a stock terminator for closing a unit of speech, and a filled epithet slot for the recurring subject. Finding one is unremarkable. Finding both, at 67 and 63 instances, distributed across 40+ chapters each, is a measurement rather than an impression — and it is the sort of measurement that cannot be talked into existence, because the phrase either occurs or it does not.

The third pattern is worth naming because it is different again: "Ask for a boon" at 25 instances is a plot slot, not a phrase slot. It marks a recurring narrative event rather than a recurring turn of speech. That the boon-grant is formulaic at the level of wording as well as structure is a small additional finding, and it connects the counting instrument to the volume's most-worked mechanism.

Keep the Claim Narrow — Here Is Exactly How Narrow

This is where pages like this go wrong, so the boundary is drawn explicitly.

What the counts establish: the text exhibits formulaic composition. Stock phrases occupy recurring slots at high frequency, distributed across the work rather than concentrated. [n≥3 PATTERN], and the pattern is arithmetical rather than interpretive.

What the counts do not establish, and must not be allowed to drift into:

• Not oral origin. Formulaic composition is consistent with oral composition and also with written composition inside a formulaic tradition, with dictation, with copying, and with deliberate archaism. Parry-Lord style arguments require more than frequency and this build does not have it.
• Not a date. Nothing about a phrase count implies when.
• Not single versus multiple authorship. A shared formulary is exactly what would let many hands produce a uniform-sounding text — so high formula density is, if anything, weak evidence against being able to detect authorship this way at all.
• Not a judgement about quality. The register is doing what it is built to do.

⚑ The number is real and the inference beyond formulaic composition is not licensed. Stating it larger would be more interesting and would be wrong.

The Second Instrument Agrees, Which Is the Point

Frequency counting is one instrument. A second, independent one was run on the same question and returned a compatible result.

The created-being formula — a god needing an agent, creating one from his own body, and issuing it a brief — occurs ten times across the corpus with four different creators (Pārvatī at ch 163, Viṣṇu at ch 174, Śiva's eyebrows at ch 189, among others), and the agent-slot is never left unfilled.³

That is a structural instrument rather than a lexical one. It counts narrative moves, not words, and it is immune to translation choices in a way phrase-counting is not.

⚑ Two instruments of different kinds, measuring different features, agreeing that the composition is slot-based. A formula that survives a change of creator is materially stronger evidence than the same formula ten times with one creator — because a repeated phrase could be one author's tic, while a repeated structure filled by four different agents is a template being used.

Neither instrument alone would carry much. Their agreement is the finding.

⚑ ENRICHED 2026-08-11 — and a third instrument agrees, on a different volume.

This page counted phrases. Volume 3's sweep counts forms — constructions with a variable slot — and the result is a different and larger object.^V1

Roughly 120 instances across five constructions, and the finding is what it does to the volume's "absolute" statements: ⚑⚑ they are overwhelmingly positions in a template, not claims.

Three consequences, each checkable.

The superlative is a slot and not a ranking. Forty-two-plus instances across all four Saṃhitās, with at least twenty distinct fillers — Śivarātri, food, truth, battle, ancestors, the pañcākṣara, dhyāna, a liṅgam, a bīja mantra, vijñāna, the ātman, a sacrifice, karma — and no two are ever reconciled. ⚑ Ch 325-6(11) stacks five in a single paragraph: no donation equal to food, no misery equal to hunger, no disease equal to hunger, no happiness equal to health, no enemy equal to anger.^V2

The inexpressibility formula is a stopping rule, not humility. Fifteen instances, and sampling where they fall shows them at chapter ends — ch 316's and ch 317's are each the final sentence of their chapter. ⚑ The speaker declares the subject inexhaustible and stops. ⚑⚑ And it is deployed by every rank: Suta, Upamanyu, Sanatkumāra, and at ch 426 Śiva himself is said to be incapable. At ch 309 it sits directly beside its own contradiction — *"I am incapable of describing the fruits. There is nothing that I cannot give in return."*^V3

And the phala-śruti's number is in the slot too. The same five-move block, with the same cohort formula — "kings, brahmanas and excellent vaishyas" — closes the Koṭi Rudra Saṃhitā at five times and the entire Shiva Purāṇa at seven times. ⚑ Same template, same reward, different count, unflagged, and the two are the only occurrences of the formula in the volume.^V4

⛔ The stretch this must not take. These five forms explain a great deal of the volume's apparent incoherence — why thirty contradictory positions can be listed without strain, why forty-two different things are supreme. ⚑⚑ They do not explain the doublets. Two hell systems, two Kauśikī origins, three Sāmbas and the Kena yakṣa told twice with the deity swapped are narrative, not formulaic, and no slot-filling account reaches them.^V5 The template finding and the doublet findings are separate and must stay so.

Where the Count Is Soft

One limitation is structural and cannot be argued away.

The object counted is an English translation. "Prostrated himself" is Debroy's rendering; the underlying Sanskrit may use several different constructions that a single English phrase collapses, or one construction that he sometimes renders differently. Either error runs in a direction that matters — collapsing inflates the count, varying deflates it.

Two things partly contain the problem.

The counts are internally comparable: whatever Debroy's rendering habits are, they are the same habits across all 134 chapters, so the distribution finding survives even if the absolute numbers do not. A phrase appearing in 41% of chapters is a fact about this translation's consistency and, at minimum, a fact about something regular underneath it.

And the created-being instrument does not depend on wording at all, which is why its agreement matters more than it would if both instruments were lexical.

⚑ The honest statement: the absolute figures are properties of this translation. The distribution and the structural corroboration are properties of the text. Anyone wanting the first as a claim about the Sanskrit needs the Sanskrit.

Implementation Workflow

You suspect a body of writing is more formulaic than it presents itself as being — a corpus of press releases, a set of AI-generated drafts, a company's documentation, a genre you are trying to learn.

Count, don't feel. Pick six to ten candidate phrases from your impression and count them. This takes minutes and it will surprise you in both directions: some phrases you were sure about turn out rare, and one you had not noticed turns out to be everywhere.

Always report the denominator. Ninety-nine instances means nothing alone. Ninety-nine in 165,512 words, one every 1,670 is a rate somebody can compare against something else.

Then check the distribution, because this is the step that decides the finding. Count the documents or sections carrying the phrase, not just total instances. High count in few sections is a fact about those sections. High count spread wide is a fact about the register — and only the second supports the claim you probably want to make.

Sort the hits by function. A stock closer (there is no doubt, at the end of the day, it is important to note) is doing something different from a filled epithet slot (industry-leading, the ever-compassionate lord) which is different again from a plot slot — a recurring event, not a recurring phrase. Finding all three is the strong result; finding only closers usually means somebody has a verbal habit.

⚑ And find a second instrument that does not depend on wording. Structural repetition — the same move made by different actors, the same section shape, the same argument order — is immune to the vocabulary problem that makes phrase-counting soft. If the lexical and structural instruments agree, you have something. If only the lexical one fires, you may have measured a style guide.

Evidence, Tensions, Open Questions

Evidence. All seven headline counts and eleven secondary ones re-run independently for this page over body lines only, chapters 171–304, 165,512 words; chapter-level distribution computed for six of them.¹ The stacked instance at ch 176 read in place.² The created-being figures carried from the corpus record and flagged as such.³

Tension 1 — the object is a translation. Absolute counts are properties of Debroy's English. Distribution and the structural corroboration are not, which is why the claim rests on those.

Tension 2 — formula density cuts against authorship detection. A shared formulary is precisely what would make many hands sound like one. This finding therefore weakens, rather than supports, any attempt to identify compositional layers by voice — which sits uneasily beside the volume's one flagged register seam, where a difference in composition was detectable.

Open question. Does the flagged Andhaka block run a different formula profile from the rest? The instrument exists and the block is identified; the comparison has not been made, and it would be a genuine test rather than a restatement.

Author Tensions & Convergences

Against The Register Seam at Andhaka: the direct tension. That page finds a block whose composition is different enough for a translator to flag it at the chapter heading; this one finds a register uniform enough that phrase frequency runs steady across 134 chapters. Both are true, and the unrun comparison between them is the obvious next instrument.

Against The One Thing the Lens Found: this is the clearest specimen of that page's rule — the working instruments measure things the author was not choosing. Nobody decides how often to write there is no doubt; it comes with the register.

Against The Boon That Cannot Be Refused: "Ask for a boon" at 25 instances across 23 chapters puts a number on that page's mechanism. The boon-grant is formulaic in wording as well as in structure.

Against The Refuge-Duty That Binds Everyone: "sought refuge" at 68 instances across 42 chapters is the highest-frequency construction tied to a specific doctrine, which is a small independent confirmation that the refuge-duty is as central as that page argues.

Cross-Domain Handshakes

Removal, Not Acquisition: The Subtractive Theory of Voice — that page argues, from the Sarasvatī Vandana, that voice is not acquired but uncovered: you remove the borrowed, the imitated and the inert, and what remains is yours.

This text is the same tradition running the opposite polarity, and the contrast is instructive rather than merely opposite.

Here the borrowed material is not an obstruction to be cleared. It is the medium. Affectionate towards his devotees sixty-three times, there is no doubt sixty-seven, an epithet slot filled almost every time the god acts — subtract the inherited phrasing from this and there is no text left, because the inherited phrasing is what makes a listener recognise what kind of utterance they are hearing and what is expected to follow.

⚑ Which suggests the subtractive theory is doing something narrower than it sounds, and the narrowing is useful. It is not a theory of writing; it is a theory of writing where individuation is the goal. In a register whose job is transmission — where the same story must be recognisable in a hundred tellings by a hundred reciters — individuation is a defect, and the formulae are the mechanism that prevents it. Two people using the same slots produce compatible texts. Two people who have each removed everything borrowed produce two texts that cannot be pooled.

So the operative question the pairing yields is not how much of this is borrowed but what is this writing for. Where the answer is "so that it survives being retold by people who are not me," acquisition beats removal, and the count on this page is what that choice looks like when measured.

One handshake, not two. The formulaic-composition literature has no page-level counterpart in the vault, and a second connection would be manufactured.

The Live Edge

Sharpest implication. Prostrated himself ninety-nine times. There is no doubt sixty-seven. Affectionate towards his devotees sixty-three. Sought refuge sixty-eight. Across 165,512 words and 134 chapters — and, crucially, spread rather than clumped: the top formula appears in 55 of 134 chapters, the epithet in 42, the closer in 41, in narrative and dialogue alike rather than confined to the hymns where stock phrasing would prove nothing. Two of the three classic signatures of formulaic composition are present at high frequency — a stock terminator and a filled epithet slot — and a third, the boon-grant, is formulaic at the level of plot as well as phrasing. A second, wholly independent structural instrument agrees: the created-being template runs ten times with four different creators and its agent-slot is never left empty, which is worth more than the phrase counts because a template surviving a change of user is not one author's tic. And then the discipline: none of this establishes oral origin, date, or how many hands were involved. A shared formulary is exactly what lets many hands sound like one — so the strongest reading of this page is that it makes voice-based authorship detection on this corpus harder, not easier.

Generative questions.

1. The Andhaka block is flagged as compositionally distinct and the formula instrument exists. Does its profile differ — and if it does not, what does that say about what the translator was actually detecting?
2. Two of three formula signatures are present. What would the third — a metrical or positional constraint on where the slots fall — even look like in a translation, and is it recoverable at all?
3. If a high formula count is what makes many hands sound like one, is there any lexical instrument that could detect layering in a text like this, or is the whole approach foreclosed?

Connected Concepts

• The Register Seam at Andhaka — uniformity against a flagged discontinuity, and the test not yet run
• The One Thing the Lens Found — why unchosen features are the ones worth counting
• The Boon That Cannot Be Refused — the plot slot, with a number on it
• The Refuge-Duty That Binds Everyone — the highest-frequency doctrine-bound construction
• The Four Doublets and the Four Disclosures — repetition at episode scale rather than phrase scale
• Removal, Not Acquisition: The Subtractive Theory of Voice — the same tradition, opposite polarity, and what that reveals about both

Footnotes