The Saṃhitā That Grew

A Book That Publishes Its Own Weights

Most old books do not tell you how long they are meant to be. This one does.

Partway through, in the opening chapter of its largest section, the Shiva Purāṇa stops narrating and reads out a table: twelve named parts, each with a number of verses attached, then the seven that survive an editorial cut with their reduced numbers. It is the sort of passage a reader skims. It is the front matter of a warehouse, printed in the middle of the aisle.

And it can be checked, because the book it describes is the book in your hands.

Add up the twelve original figures and you get exactly one hundred thousand. Add up the seven surviving figures and compare them against the parts of the printed text they name, and the total comes out within half a percent. Five of the seven land within a few percent of their stated size.

Two do not. One section is forty-six per cent over its published weight. Another is twenty-three per cent under. The scale is right; the distribution is wrong. Something is sitting in the wrong part of the book.

The Original Twelve Come to Exactly One Hundred Thousand

The passage is at ch 388-8.1(1) — cited that way, and never as "ch 388," because this edition uses that number for two different chapters in two different Saṃhitās.¹

Scripted, and exact.¹ Five of these — Vaināyaka, Mātṛ, Rudraikādaśa, Sahasrakoṭirudra, Dharmasaṃjña — no longer exist, having been cut by Vyāsa in the abridgement the text itself discloses.²

⚑ Worth pausing on before the interesting part: the text's arithmetic here is perfect. Twelve figures summing to a round hundred thousand is not the behaviour of a corpus that cannot count, and it is a useful control on everything that follows. The failures in this volume are localised.

⛑⛑ And the control has a condition on it, found while writing this page. ⭐ The same table appears in Volume 1, at ch 9-2(2), and it is not the same table. Two entries differ: "Kailasa had 3,000 shlokas and Shatarudra half that number" against this volume's "Kailasa has six thousand shlokas, Shatarudra has three thousand."¹² Both passages then assert an aggregate of one hundred thousand. Volume 1's figures come to 95,500. Volume 3's come to 100,000. Unflagged in both volumes; no note in either range remarks that the tables differ or that one of them does not total.

⛔ What that costs this page, stated exactly. The control still holds for the table this page uses, which is Volume 3's, and which is exact. What it can no longer be is a control on the corpus — because the corpus prints two versions of its own budget and only one of them adds up. ⚑ And the direction is suggestive without being determinable: the two disputed figures are precisely doubled here, and the doubling is what makes the total come right. "Half that number" — a value expressed as a ratio rather than a figure — is the likeliest site of the damage on either reading. [n=1 OBSERVED].

The Abridged Seven Do Not

The same chapter then gives the abridged sizes. Set them against the translator's own count of the Sanskrit text he worked from, stated in the Introduction:³

The total is almost exact. The distribution is not.

That combination is the whole finding, and it rules out the boring explanation. If the printed text simply came from a fuller or thinner recension than the one ch 388-8.1(1) describes, every row would drift the same way and the total would drift with them. Instead five rows sit still, one row is enormous, one row is small, and the two errors very nearly cancel.

⚑ Errors that cancel are not errors. They are a transfer. Something of roughly the right size is in the Umā Saṃhitā and is not in the Vidyeśvara.

Analytical Case Study — Where the Extra Eight Hundred Shlokas Are Sitting

The Umā Saṃhitā is fifty-one chapters, ch 315–365. It is also, and this was established by a completely different instrument before the shloka table was ever added up, the only part of this volume in which material is demonstrably imported from other texts.⁴

Four separate blocks, in one Saṃhitā:

• Ch 338, the Pañcacūḍā dialogue. Translator-flagged, in advance: "In the Mahabharata, she had a conversation with Narada on this issue. Indeed, the next chapter reproduces those sections from the Mahabharata, almost verbatim."⁵
• Ch 352, the Satyavrata material. The body reads "O brahmanas!" in the wrong mouth, and the note explains: "This is an inconsistency, because Vaishampayana is speaking to Janamejaya, though the conversation is being repeated to the brahmanas." Vaiśampāyana narrating to Janamejaya is the Mahābhārata's own outermost frame — running inside this Purāṇa, with the Purāṇa's own vocative leaking into it three times in one chapter.⁶ ⚑ And the body states the frame itself: "This is exactly what Vaishampayana was asked by Parikshit's son."
• Ch 354–356, the ancestors and the seven hunters. Five frames deep, with Mārkaṇḍeya speaking to Bhīṣma and the vocatives breaking three times, all three flagged: "Since Markandeya is speaking to Bhishma, this is an inconsistency."⁷
• Ch 359–364, the Devī material — the dispossessed king Suratha and the vaiśya Samādhi, the sage Medhas, Madhu and Kaiṭabha from Viṣṇu's ear-wax, and Mahiṣāsura killed half-emerged, in that order. That is the Devī Māhātmya's frame, its three episodes and their sequence. [PLAUSIBLE — needs corroboration]; the collation has not been run and this page does not claim it has.⁸

Two wholly independent instruments — the text's own published budget and the translator's frame-flags — name the same Saṃhitā. One of them is arithmetic on a table; the other is a census of misfitting vocatives. Neither was built to find the other.

⚑ And the shape of the joins is diagnostic in itself. When this text absorbs a block that arrives with its own speaker and listener, the host's vocatives leak into the graft — the reciter is speaking inside the outer frame while transmitting the inner one. The leak is not a scribal slip. It is a seam.⁴

The Third Instrument, and It Is the Translator's

There is a third leg, and it is the one that makes the direction of the transfer legible rather than merely the fact of it.

Attached to ch 388-8.1(1)'s own heading — the heading of the chapter that publishes the budget — is note 1776, in full: "This chapter repeats bits from Vidyeshvara Samhita."⁹

Vidyeśvara is the Saṃhitā that came up four hundred and sixty shlokas short. And the Vāyavīya's opening chapter is repeating it.

⚑ So: material that ch 388-8.1(1) assigns to the Vidyeśvara is physically located elsewhere in the book, and the translator says so in a note attached to the assignment itself. Three instruments, three methods, one conclusion about which sections have moved.

⛔ What this does not establish. It does not establish that the eight hundred and thirty-nine shlokas in the Umā came out of the Vidyeśvara. Nothing connects the two anomalies except that they are opposite in sign and roughly similar in magnitude, and note 1776 concerns a different Saṃhitā's opening chapter. Two independent displacements are as consistent with the evidence as one transfer, and this page does not choose between them.

Expansion by Absorption, Stated at Its Real Strength

Set the Umā's behaviour beside the other two Saṃhitās this volume contains and a second, different phenomenon appears — which matters, because merging them would be the easy error.¹⁰

The Umā grew. The Kailāsa and the Vāyavīya did not grow — they simply never agreed.

⚑ That distinction is precisely what the shloka budget buys, and it could not have been drawn from the contradictions alone. A text with two incompatible tattva schemes might be a text that swallowed something, or a text assembled from two intact books that were never reconciled. The budget separates the two cases, because absorption leaves a size signature and independence does not.

⛔ [n=1 OBSERVED] on the magnitude. Shloka counts vary across manuscript traditions; Debroy's figures are his own and unverified by this build; this is one measurement. What is solid is the direction, the concentration, and that two instruments agree. No rigour-register language attaches.

The Falsification Test, Runnable

The reading has a condition that would break it, and it can be checked without any new source.

If the Umā Saṃhitā grew by absorbing narrative blocks that arrived with their own frames, and if the Vidyeśvara correspondingly shrank, then the Vidyeśvara Saṃhitā should show no imported frames at all — a shrunk section has not been receiving material.

Volume 1 contains the Vidyeśvara Saṃhitā, twenty-five chapters. The frame-violation instrument runs on it in an afternoon: sweep for misfitting vocatives, unmarked interlocutor switches, and named external frames, and count.

• A null result confirms the shrinkage reading, and converts a two-instrument convergence into a three-instrument one.
• Imported frames in the Vidyeśvara falsify the transfer reading, and would leave the +46% standing as a fact about the Umā with no account of where the material came from.

⚑ The test is stated here before it is run, deliberately. This volume's build has already had one script-verified structural finding collapse on contact with an endnote, and the discipline that caught it was writing the falsification condition down first.¹¹

Implementation Workflow

You have a system that publishes its own specification — a codebase with a manifest, an organisation with a headcount plan, an accounts pack with a stated allocation — and you want to know whether the specification and the thing are the same thing.

Total it first, and then stop looking at the total. You add the figures and the sum comes out right, and the temptation is enormous to file the check as passed. It is the wrong check. A correct total across a wrong distribution is exactly what a transfer looks like, and it is the single most common way an internal move survives an audit: the money is all still there.

Find the rows that move and ask what they have in common. Not what they are, but what has happened to them. Here two rows move and five sit still — and the one that grew is independently, by an unrelated method, the only part of the book containing material demonstrably written elsewhere. That coincidence is the finding. The arithmetic on its own would only have been a curiosity.

Then look for the join, not the content. You are trying to date an insertion in a document nobody signed. Do not read for style — style arguments never settle. Read for the places where the host's habits leak into the guest's material: a form of address that belongs to the outer conversation appearing inside the inner one, a house convention surfacing in a quoted block, a naming scheme from one team showing up in another team's file. The leak marks the seam because the person doing the joining was working in their own idiom while carrying somebody else's.

And write the falsification condition before you run the test. Say out loud what result would make you abandon the reading. Then run it. An instrument with no stated failure condition will eventually return something, and you will not be able to tell whether it found anything.

Evidence, Tensions, Open Questions

Evidence. Ch 388-8.1(1), lines 993–1005, read in full and both tables extracted; the original twelve summed by script to exactly 100,000; the abridged seven set against the Introduction's own count of the printed Sanskrit at line 119.¹³ The four import blocks located independently, three of them translator-flagged at their own note lines.⁵⁶⁷⁸ Note 1776 read at its own line.⁹ The Kailāsa–Vāyavīya technical divergences counted and cross-checked.¹⁰

Tension 1 — the two anomalies may not be one event. The page reports a +839 and a −460 that nearly cancel, and declines to assert a single transfer between them. ⚑ The honest position is that the total's accuracy is evidence that the book as a whole is the book the budget describes, and the rows' inaccuracy is evidence that its internal boundaries are not where the budget puts them. Whether that happened once or twice is unknown.

Tension 2 — Debroy's shloka counts are the measurement instrument and they are not independent of him. He counted the printed Sanskrit; nobody has re-counted it here. If his figures carry a systematic error the whole table moves. ⚑ The mitigating fact is that a systematic error would not produce five accurate rows and two wild ones.

Tension 3 — the Devī identification is unconfirmed and the page depends on it least. Strike ch 359–364 from the import list entirely and the Umā still carries an announced verbatim Mahābhārata reproduction, a named Mahābhārata frame, and a three-leak Mārkaṇḍeya frame. The finding survives its weakest leg being removed, which is the property it should have.

Open question. Does any other Purāṇa publish a comparable internal budget against which its own printed text can be measured? If the practice is common, this becomes a general method for locating displaced material in the corpus. If it is rare, this volume is a lucky instrument and the method does not generalise.

Author Tensions & Convergences

Against The Five Lost Saṃhitās: that page records the abridgement — five Saṃhitās cut, fifty-two thousand shlokas named as gone. This page supplies the table it was working from and confirms its figure: the five excised sections sum to 8,000 + 8,000 + 13,000 + 11,000 + 12,000 = 52,000, exactly. The two pages are the same passage read for different things — one for what was removed, one for whether what remains sits where it says it sits.

Against The Vocative That Does Not Fit: that page's long-standing verdict was that the frame-violation instrument "never collapsed — it simply never did anything," on a record of five instances in Volume 1, all translator-flagged, none found unaided. This page is what the instrument was actually for. It is not a detector of composite composition; it is a map of where composite composition already is, and it only becomes legible when a second, unrelated measurement is laid over it. ⚑ An instrument that returns nothing on its own terms can still be correct about where — which is a distinction the vault did not previously have.

Against The Apparatus Is the Translator's: the awkward convergence, and it has to be stated. Two of the three legs of this finding are supplied by the translator — his count of the printed Sanskrit, and his note flagging that the chapter repeats the Vidyeśvara. The page that catalogues his interventions and the page that depends on his measurements are the same build. There is no way around this; there is only saying it. The mitigation is that the third leg, the frame-clustering, is a pattern across his flags rather than any one of them.

Cross-Domain Handshakes

Memory Malleability and Perception Construction — that page's claim is that a remembered episode is rebuilt at each recall from fragments and expectation, and that the rebuild leaves no marker distinguishing it from the original.

The shloka budget supplies something that literature does not usually have and this page can offer it back: a case where the reconstruction has a published weight.

The reason reconstructive memory is so hard to catch in the act is that there is no ledger. A person's account of an event has no stated length, no manifest, no total to which the parts must sum — so an absorbed detail from another occasion is indistinguishable from a retained one, and the only detectable signature is content, which is exactly the kind of evidence that never settles an argument. Here there is a ledger. The text says how big each of its parts should be, and one part is half again as big as its entry, and the excess sits precisely where material from other books demonstrably arrived.

⚑ The transferable move is not "texts absorb things," which is obvious. It is look for the invariant the system publishes about itself, and check the distribution rather than the total. Absorption almost never changes the aggregate — a memory does not get longer, an account does not run over time, a budget does not miss — because the aggregate is the thing under supervision. It changes which bucket holds what. So the diagnostic for any reconstructive system with a published constraint is: find the constraint, satisfy yourself the total holds, and then treat the total's holding as the reason to go looking, not as the reason to stop.

That is a genuinely two-way exchange. The psychology page explains why the total holds — the aggregate is what the system is optimising to keep plausible. This page shows what it looks like when the internal partition is auditable anyway.

The Live Edge

Sharpest implication. The Shiva Purāṇa publishes a table of its own sections' sizes; the twelve original figures sum to exactly one hundred thousand, and the seven abridged figures come within half a per cent of the printed text overall. But one Saṃhitā runs forty-six per cent over its published weight and another twenty-three per cent under, while the other five sit still — so the scale is right and the distribution is not. And the Saṃhitā that is over is, on a completely independent instrument built for another purpose, the only part of this volume carrying material demonstrably imported from other texts: an announced near-verbatim Mahābhārata reproduction, the Mahābhārata's own Vaiśampāyana frame with the Purāṇa's vocatives leaking into it, a five-deep Mārkaṇḍeya frame that breaks three times, and the Devī Māhātmya's three episodes in order. A third leg is the translator's own note on the budget-chapter's heading: it repeats the Saṃhitā that came up short. The Umā grew by absorption; the Kailāsa and Vāyavīya did not grow and simply never agreed — and that distinction is what the budget buys and the contradictions alone could never have supplied.

Generative questions.

1. The total is right and the rows are wrong. What would have to be true of a transmission history for a text's aggregate size to be better preserved than its internal boundaries — and is that what an abridgement to a stated target actually does to a book?
2. Volume 1's Vidyeśvara is the falsification test and it has not been run. If it returns imported frames, what remains of this page?
3. The frame-leak is a join, and every join found so far is in narrative material. Would the same instrument detect a ritual graft, where there are no speakers to leak?

Connected Concepts

• The Five Lost Saṃhitās — the same table, read for what was cut
• The Vocative That Does Not Fit — the instrument, and what it turned out to be for
• The Apparatus Is the Translator's — two of this page's three legs are his
• The Four Doublets and the Four Disclosures — the machinery for announcing repetition, and where it is not used
• Memory Malleability and Perception Construction — absorption without a marker, and what a published ledger adds

Footnotes