Somewhere in the middle of the second volume there is a roll-call of the saved. Dozens of names, each with a one-line outcome attached: this one worshipped and obtained the supreme destination, that one was a sinner and was freed anyway. It reads like an index — the sort of list that exists because the stories exist somewhere and you are being reminded of them.
Five of those names were checked against the rest of that volume. Each occurred exactly once, inside the list, and was never narrated anywhere. The build recorded it as an open question and carried it forward: do these people have stories in the third volume?
Three volumes are now read. The answer is no. And what the apparatus does when asked for the missing stories is the actual finding, because it does two different things and both of them fail.
Whole-file sweep of Volume 3 for the five names:1
| Devotee | Hits in Volume 3 |
|---|---|
| Kalāvatī | 0 |
| Śucivrata | 0 |
| Citravarmā | 0 |
| Vimarśana | 1 in the body, plus one endnote |
| Śāradā | 1 in the body, plus one endnote |
⚑ The finding is not that two of five turned up. It is where they turned up. Both reappear inside another litany — not in a narrated episode, not with a scene, not with a speech.
Ch 316-6(2), in full on Vimarśana: "O Krishna! In earlier times, King Vimarshana was devoted to Shiva. Through Shiva's favours, he obtained a supreme destination."2
Line 1193, in full on Śāradā: "Sharada was a widow, but you acted so as to give her husband back to her."2
That is the whole of it, on both. A name recurring inside a second list is not a story; it is the same index entry printed twice.
Asked what Śāradā's story is, the apparatus supplies endnote 1857, which reads in full:
"Narrated in Volume II, Chapter 195-3.5(25)."3
That is the litany. It is the very list the earlier build flagged as containing five names that occur exactly once, inside it, and are never narrated. The cross-reference resolves to the passage that created the gap.
⚑ It is worth being precise about why this is a finding rather than a slip. The note is correct. Śāradā is mentioned at 195-3.5(25). Everything the apparatus says is true, the reference is properly formed, and a reader who follows it will arrive exactly where they were sent. What they will not find is a narrative — because there is no narrative, and the note has no way to say so.
A circular cross-reference is what a citation apparatus produces when the thing being cited does not exist. It cannot report an absence; it can only report the nearest presence, and here the nearest presence is the same absence in a different volume.
[n=1 OBSERVED], and it is the sharper of the two items on this page.
Where the apparatus cannot point, it supplies. Endnote 1143, on Vimarśana:
*"Vimarshana was a king of the Kirata kingdom. Inadvertently, in the form of a dog, he died in front of a Shiva lingam and was rewarded in successive births."*4
A kingdom, a manner of death, an animal form, an inadvertence, and a sequence of rebirths. The Shiva Purāṇa nowhere says any of this, in any of its three volumes.
⛔ Stated at its real strength: the story is not fabricated. It is almost certainly drawn from the wider Purāṇic tradition, where such a story plainly exists, and Debroy is supplying what a reader would want. What is missing is any marker that the note has left the text. It is printed in the same typeface, in the same voice, in the same numbered sequence as the notes that gloss a word or record an emendation.
⚑ And the content is not neutral. "Inadvertently, in the form of a dog" — an accidental merit, earned by an animal, without intention — is precisely the doctrine this volume argues about at length in its own voice.5 A note that arrives unmarked from outside is now supplying evidence in an argument the text is having.
Two further cases in this volume show the same import behaviour doing heavier work, and they are worth having on the same page because they establish that note 1143 is not an anecdote.
Note 1379 attaches to ch 350-6(36), which narrates Sudyumna as born female and granted manhood as a reward. The note supplies a different story entirely — "Sudyumna had wandered into a grove where Shiva and Parvati were amusing themselves in private… According to the standard story, that is how Sudyumna became Ila" — which is a curse, in the opposite direction, and matches the version this volume gives at ch 308-5(4).9 The apparatus restates the contradicted version as standard, on the page that contradicts it, without saying the two conflict.
Notes 1979 and 1980 are two bare names, four words apart in the body. Ch 437-8.2(15) says "Though she was devoted to her husband, Atri's wife abandoned her husband", and then, with nothing in the English to mark a change of subject, "She was not engaged to do this by her guru… and obtained the Pandavas as her husbands." Note 1979 reads "Anasuya." Note 1980 reads "Droupadi."10
⚑ He supplies both identifications and never states that the subject changes mid-passage. A reader who clicks both notes gets it. A reader who clicks neither reads one woman throughout — and the sentence they read is that the tradition's exemplar of devotion to her husband abandoned him and married five men.
⚑ Which is the general shape: the body reads smooth because the apparatus is carrying the load, and the smoothness is what makes the load invisible.
Put the two together and the object comes into focus.
Three volumes carry devotee-litanies — a scripted census of the third volume alone finds 133 instances of the form [Name] … obtained / became / was freed, with the three principal litanies carrying 85 named-figure-with-outcome sentences between them.6 Read as an index, that is an enormous apparatus pointing at a body of exemplary lives.
The body of lives is not there. At least five of the names index nothing at all; and where a reader goes looking, the apparatus either invents or points in a circle.
⚑ The honest conclusion is about what a litany is, and it is not a summary. It is a form — a slot with names in it, doing the work of establishing that many people have been saved, and requiring no stories to do it. [n=2 SUGGESTIVE] on the two recurrences; the circularity is [n=1 OBSERVED]. Neither number licenses a claim about all 133.
⛔ And a caution the page owes itself: some of these names are narrated. Cañculā's story is told at length in Volume 1; Gurudruha's occupies a whole chapter here.5 The claim is not that the litanies are hollow. It is that the litany form does not require its entries to be backed, and at least five of them are not.
This is the item with the widest reach, and it is a rule rather than an observation.
This vault cites Debroy's endnotes as evidence about this text. Note 1143 is evidence about the tradition, presented in the same apparatus, in the same voice, with no marker.
Before any page cites a note, establish which of four things it is doing:
⚑ Four different acts, one typeface, one numbered sequence. The reader distinguishes them only by reading each note and asking.
You are working with a reference apparatus — footnotes, a bibliography, a linked knowledge base, a set of internal tickets — and you want to know whether the references resolve to anything.
Follow one all the way down. Not to the next node: to the bottom. Pick an entry that the system treats as well-supported and walk the chain until it terminates in a primary object or comes back to somewhere you have been. The failure mode you are looking for is not a broken link — a broken link announces itself. It is a working link that resolves to another pointer, and then to another, and eventually to the same list you started from.
Distinguish an index from a summary. They look identical and they carry opposite guarantees. A summary is downstream of content and implies content exists; an index is a form and implies nothing. Test which you have by picking three entries at random and asking what they point at. If two of three point at other lists, the object is a form.
And check what the annotator does when the source is silent. This is the diagnostic that does the most work, because it exposes the annotator's default rather than their accuracy. Some fill the gap from general knowledge, unmarked. Some point at the nearest thing. Very few write not present. ⚑ Whichever they do, they will do it consistently, and knowing which one tells you how to read every note in the apparatus — including the ones that happen to be right.
Evidence. Whole-file sweep of Volume 3 for all five names, with every hit read in place.1 Both body occurrences read in full at their lines.2 Notes 1857 and 1143 read at their own lines and quoted in full.34 The 133-instance litany census run by script across the body.6 The hedging census run across all 2,124 notes.7
Tension 1 — the two notes fail in opposite directions and cannot both be the apparatus's habit. One invents, one points in a circle. ⚑ The reconciliation this page offers is that they are the same default under different conditions: supply the nearest available content, where "nearest" is a remembered story in one case and a cross-reference in the other. That is a reading, not a demonstration, and it is offered as one.
Tension 2 — the negative result is a negative on a name sweep, and names are unstable in this edition. Note 1139 discloses that proper names are amended across volumes to harmonise them.8 A devotee narrated under a variant spelling would not be caught. The three zero-hit names are therefore zero hits on those forms, which is weaker than absence and is stated as such.
Open question. Three of the five have zero occurrences in Volume 3. Do their stories exist anywhere in the wider Purāṇic corpus, as Vimarśana's evidently does? If they do, the litany is indexing a tradition rather than a text — which would be a different and more interesting finding than a gap, and would make the apparatus's imports the correct behaviour rather than an unmarked one.
Against The Apparatus Is the Translator's: the same object seen from its other side. That page catalogues what the apparatus changes; this one catalogues what it supplies. ⚑ The two acts have the same structure — a decision printed in the same typeface as the thing decided about — and this page's version is the harder to catch, because a substitution at least replaces something that was there and an import replaces nothing.
Against The Corpus of the Disqualified: the same 133 litany instances read for who is in them — demons, three courtesans, a caṇḍālī, hunters, a tiger, a man who died as a dog, two people who changed sex — against access rules that exclude all of them. This page reads them for whether they are backed. Both findings are true of one object and neither weakens the other: the litany is a form, and the form is filled with the disqualified, and the second is more interesting precisely because the first shows the filling was a choice.
Against The Nested Narrators: a litany is the most compressed possible instance of that page's structure. Every entry is a claim with a narrator and no scene — somebody inside the frame asserting an outcome nobody has to have witnessed. There is no position in the book from which an entry could be checked, and no entry ever is.
When the Crowd Is Wrong Together — that page's claim is that a manufactured testimonial montage "borrows the shape of organic consensus without the substance that made organic consensus trustworthy," and that the shape is what does the persuading.
The litany supplies a limiting case that the social-proof literature mostly does not have: proof whose instances are not checkable and were never meant to be.
Standard social proof runs on instances a listener could in principle verify — the queue outside the restaurant, the number of purchases, the neighbour who switched. The persuasive force is usually explained by inference from the aggregate. Here the aggregate is the only thing there is. A hundred and thirty-three names with outcomes attached, five of which are known to index nothing, and no mechanism anywhere in the text by which a listener could tell a backed entry from an unbacked one. And it plainly works — the litanies are among the most repeated structures in three volumes.
⚑ What that isolates is worth having: the persuasive unit is the enumeration, not the cases. What the listener takes away is many, various, including people like me — and every one of those properties is carried by the list's shape rather than by any entry. Which predicts something checkable in ordinary settings: that testimonial sets are optimised for spread and count rather than depth, that the disreputable inclusion is doing more work than the flagship one, and that a set which invites you to examine any single case is a different rhetorical object from one that invites you to weigh the pile.
And it runs back the other way. The reason nobody in a thousand years appears to have asked who Kalāvatī was is that an index is not experienced as a set of claims. It is experienced as a magnitude. That is the mechanism the behavioural page has, applied to a text, and the text is what shows how far it can be pushed: all the way to entries with nothing behind them, in a document whose own translator, asked, points back at the list.
Sharpest implication. Five devotees named in a Volume 2 litany occur exactly once each, inside it, and are never narrated. A whole-file sweep of the concluding volume closes the question: three have zero occurrences, and the two that recur do so inside another litany, in a single sentence apiece, with no scene attached. Asked for the missing stories, the apparatus fails in two different ways. Endnote 1857 reads, in full, "Narrated in Volume II, Chapter 195-3.5(25)" — which is the litany, the very list that created the gap; the reference is correct, properly formed, and resolves to nothing, because a citation apparatus cannot report an absence, only the nearest presence. Endnote 1143 supplies Vimarśana's whole story — a Kirāta king who died inadvertently in the form of a dog before a liṅgam — which the Purāṇa nowhere contains, unmarked as external, in the same typeface as every other note. So a form that reads like an index into a body of exemplary lives turns out to index, at least five times, nothing at all — and the rule that follows reaches every page in this vault: before citing a note, establish whether it is commenting on this text, importing from the tradition, recording an emendation, or hedging. Four acts, one typeface.
Generative questions.