Śālva is losing. So he uses what the flying city was built for.
He produces an illusion — Vasudeva, Kṛṣṇa's father, bound and brought forward — and beheads it in front of him.1
And Kṛṣṇa grieves.
The volume narrates the grief. He is described as an ordinary man would be described, momentarily undone by the sight of his father's head coming off.
And then, without a paragraph break, the narration stops being narration:
"There are some rishis who are inconsistent in their reasoning. Indeed, they do not remember what they themselves have said and are contradictory in their statements. Sorrow, delusion, affection and fear are created by ignorance. How can these exist in him?"1
⚑ Read what has just happened. The teller has interrupted his own story to accuse the transmitters of the story of incoherence — and the passage he is attacking is the one he has just delivered.
Not a rival school. Not a heretic. Some rishis — the people the whole text depends on for its content, in a corpus whose central boast is the reliability of its own chain of custody.
The accusation is oddly specific, and its specificity is why this is more than a doctrinal aside.
"They do not remember what they themselves have said." That is a charge of internal inconsistency across a body of transmitted material — not error, not ignorance, but failure to hold one's own prior statements in view.
"They are contradictory in their statements." The same charge, stated as a property of the corpus.
Then the argument: sorrow, delusion, affection and fear are products of ignorance; he is not ignorant; therefore he cannot have them.
⚑ The argument is sound and it is aimed at the sentence immediately above it.
⚠ And note the register. The narrator does not say this passage has been misunderstood or the grief is apparent rather than real. He says the people who transmit such passages contradict themselves. He does not withdraw the grief. He leaves it standing and calls its source unreliable.
There are three repairs available and all three are wrong here, and naming them is the only way to keep the finding intact.
Repair one: he was only pretending. ⚑ The volume has a working phrase for exactly that, uses it twice in one chapter twenty chapters earlier, and does not use it here. At 10(57) it says "Though he knew the truth" and then "they imitated the behaviour of the world of men." It can say this when it means it. It does not say it.
Repair two: the narrator is correcting a mistake. He is not. A correction replaces the erroneous statement. This one leaves the grief on the page and impugns the transmitters instead — which is not what you do to a sentence you are retracting.
Repair three: the two passages come from different hands and the seam is accidental. ⚠ Possibly true and unavailable as an argument here. This build has no access to the recensional history, and asserting a compositional seam from an English translation would be exactly the fabrication the vault's rules forbid. What can be said is what is printed: two statements, consecutive, incompatible.
⚑ Per CLAUDE.md's first hard constraint, the contradiction is preserved. The vault does not get to pick the verse it prefers.
The grief at 10(77) is not the volume's first, and the earlier one is the strongest evidence that the 10(77) denial is bounded rather than general.
Sixty-five chapters back, at 10(12), Kṛṣṇa is about to deal with Agha and the text says:
"He was filled with compassion, but also surprised at what destiny had done."2
⚑ Surprise. Not on the narrator's list of four — sorrow, delusion, affection, fear — but plainly in the same family, and nobody interrupts. No ṛṣis are accused. The clause passes without remark.
So the volume attributes an interior state to him early, unremarked; attributes another late; and rebuts only the late one. ⚑ The denial at 10(77) does not cover the material the volume has already printed, which is precisely the charge the narrator is levelling at everybody else: they do not remember what they themselves have said.
⚠ Stated at the register the sample carries: two instances, one rebutted and one not, in one volume. No claim about the corpus.
You write something down and then, on the next line, explain that people who write such things are unreliable. Notice that you have not deleted anything, and that the reader now holds both.
Somebody attacks a source's consistency in the middle of using it. That is not hypocrisy — it may be the only honest option available when you cannot verify and cannot omit.
You have a phrase that would resolve the difficulty and you have used it before. If you do not reach for it here, ask yourself what you would be conceding by reaching for it.
Two statements sit next to each other and cannot both be right. The temptation is to decide. The discipline is to record that they are adjacent and let the next person see the join.
Work out what the volume gains and what it spends, because the trade is unusual.
What it gains: doctrinal integrity. The position that the supreme is untouched by sorrow, delusion, affection and fear is defended, forcefully, at the exact moment a story has violated it.
What it spends: its own witnesses. ⚑ The only way to defend the doctrine here is to say that the people who transmit the stories are inconsistent — and this text is nothing but transmitted stories, arriving through a named chain: Śuka from Vyāsa, and behind them Nārada, and behind him a lineage the Twelfth Skandha gives three incompatible ways.
So the interruption does not simply protect a doctrine. It tells you how to read the rest of the book, and what it tells you is: the narrative layer and the doctrinal layer can disagree, and when they do, the narrative layer is the one with the credibility problem.
⚑ That is a genuinely radical instruction and the volume issues it in passing, once, and never applies it again. Nothing later says and this is another of those inconsistent passages. The tool is produced, used on a single verse, and abandoned.
Set the interruption beside the text's other self-auditing moments and it stands out by being the only one that attacks a source.
At 10(12) the narration halts so Parīkṣit can raise a chronology objection — the deed happens at five and is reported at six — and Śuka answers it as doctrine.
At 10(29) and 10(33) Parīkṣit raises the morality of the rāsa, twice, and gets two different answers, one by precedent and one by a two-tier rule.
At 10(75) Parīkṣit cites a version he already holds — "that is what we have heard" — and asks for the discrepancy to be explained.
⚑ In every one of those, the objection comes from the listener and the teller answers it. The apparatus works: question in, answer out, both printed.
Here the objection comes from the teller, about the teller's own material, and no answer is given. There is nobody to reply. ⚠ It is the one self-audit in the volume with no second party, and that is why it does not close.
Which is worth pressing on, because Parīkṣit is present. He is sitting there. He has interrupted four times already in this Skandha, twice on morality and once to check the reciter against a version he had heard elsewhere. He is exactly the person who would ask.
⚑ And on the single occasion the teller impugns his own sources, the listener says nothing.
⚠ Bounded: no follow-up found at 10(77) in the range read, and no claim about intent. But the volume has spent the whole Skandha establishing a listener who audits, and the one moment that most obviously calls for an audit is the one it passes over. The apparatus that works everywhere else is present and idle, which is a different and stranger fact than the apparatus simply being absent.
The strongest support is the adjacency. The grief and the denial are not chapters apart or Skandhas apart. They are consecutive, which forecloses any reading in which the volume simply lost track.
A second support. The vocabulary of the accusation — they do not remember what they themselves have said — is about a corpus's internal consistency, not about a doctrinal error. That is a bibliographic charge, in a volume whose count failures are all bibliographic.
The tension, preserved. ⚑ The volume both narrates its God's grief and denies he can have it, in consecutive sentences, and calls the ṛṣis who say otherwise inconsistent. Held exactly as printed. Neither statement is deleted and no reconciliation is adopted, including the one the text itself supplies elsewhere.
A second tension. The denial is issued once and never applied again, in a volume with plenty of further candidates. A hermeneutic tool used on one verse and dropped is either a slip or an admission, and the text gives no way to tell.
Open question. The narrator accuses unnamed ṛṣis of not remembering their own statements while the volume itself gives its transmission chain three incompatible ways across one Skandha. ⚠ Is the charge self-aware? Not determinable from the range read, and no claim is made.
Set against He Went to a Funeral for People He Knew Were Alive, this is the pairing that does the most work on the page. ⚑ That chapter establishes that the volume has an explicit category for staged emotion, uses it twice in one chapter, and marks it each time — they imitated the behaviour of the world of men. The existence of that phrase is what makes its absence here a finding rather than an omission.
Against They Read the Signs Correctly and Walked In, the two pages divide 10(12), and this page takes the clause that one does not need: surprised at what destiny had done. An earlier interior state, unremarked, sixty-five chapters before the rebuttal.
Against He Did Not Come to Visit. He Came to Check., the contrast is structural. Every other self-audit in this volume has two parties — a sceptic and a respondent — and closes. This one has one party and does not.
Against A Light Comes Out of the Body and Goes Into the Man Who Killed It, the same chapter-neighbourhood shows the volume perfectly willing to state a mechanism flatly and cross-reference its own instances. ⚑ It knows how to mark a category. Here it marks a category and puts its own narration inside it.
Declared Biases as Historian's Ethic — the historian states his own commitments and limits inside the work, on the ground that a reader who knows where the writer is standing can correct for it, and that concealed position is the greater failure.
The interruption at 10(77) is that ethic performed on somebody else's behalf. The narrator declares a limitation — of the transmitters, in the middle of transmitting — and leaves the reader to correct for it.
⚑ What neither field gives alone: the historian's declaration is about the writer and this one is about the sources, which makes the writer's own position unstated and unreachable. MacCulloch tells you where he is standing. Śuka tells you the ground under the story is soft and says nothing about where he is standing on it — and he is inside the same chain he has just impugned, at one remove.
Which produces the sharper version of the problem. A declared bias is correctable because you know its direction. A declared unreliability in one's own sources, issued selectively and applied once, gives the reader no direction at all — only the knowledge that some unspecified portion of what follows is subject to the same charge, with no marker on which portion. ⚠ The volume never marks another.
Authority: Institutional Override — an institutional claim outranks an individual one. The person on the spot may have the observation, but the institution has the standing, and where they conflict the institution's version is what enters the record.
The interruption is that override run inside a single text. The narrative layer has the observation — a man saw his father beheaded and grieved. The doctrinal layer has the standing — the supreme cannot be touched by sorrow. And the doctrinal layer overrides, in the narrative layer's own voice, mid-sentence.
⚑ The insight the pairing produces: an override normally silences the overridden claim, and this one cannot. The grief has already been printed. The doctrine cannot delete it, only contest it — so the record ends up carrying both the observation and the ruling that the observation's source is unreliable.
And that is why this is the volume's most important structural passage after its missing chapters. ⚠ Most institutional overrides are invisible: the overridden version simply never enters the record. Here the mechanism is visible because the text is both parties, and what it exposes is that the doctrinal layer's authority rests on a claim about the narrative layer's reliability — a claim the doctrinal layer has no independent way to establish, since everything it knows arrived through the same channel.
Sharpest implication. The volume narrates its God grieving over his father's beheading and then, in the next sentence, attacks the ṛṣis who transmit such passages as people who "do not remember what they themselves have said" — arguing that sorrow, delusion, affection and fear cannot exist in him. It does not delete the grief. It has a phrase for staged emotion, used twice twenty chapters earlier, and declines to use it. And sixty-five chapters back it printed surprised at what destiny had done with no interruption at all.
Generative questions.