Two beings in this corpus do something at a god's instigation and the outcomes could not be further apart.
The first is Kāma. He is sent to shoot Śiva, does it, and is incinerated on the spot — before the Maruts can finish saying please pardon.
The second is Arihat. He is created to write a scripture designed to destroy a civilisation's dharma so that its population becomes killable. Three cities burn because of what he does. And he is fine. Not forgiven afterwards — fine from the start, because Viṣṇu told him so before he began:
"Since this is because of my command, no sin will attach to you."1
One sentence, issued in advance, and the worst operation in the volume costs its operative nothing.
So the rule this text runs on is not the impaired are punished and it is not agents bear the cost. Something narrower is going on, and once you see it, a dozen otherwise arbitrary verdicts line up.
Liability attaches to what was undisclosed and to what was unauthorised. It does not attach to what was documented in advance — however bad the documented thing is.
Take the cases where somebody pays, and look at what they were not told.
The six wives (ch 152). Six of the seven sages' wives come out of the water cold, go to the fire, and drops of Śiva's semen enter through their pores. The text names their condition before it names their fault: they "were deluded by Shiva's maya." Nobody informed them. And the phrase that follows is the whole mechanism in six words — *"Those six wives saw that they had transgressed."*2 They discover their own guilt after the fact, in a state they did not choose and were not warned about.
Arundhatī is the control, and she is in the same scene. She alone is "instructed by Girīśa" — told in advance — and she alone is not deluded. She tries to hold the others back and fails.2 The variable that separates her from them is disclosure, and the text prints it without comment.
Tripura's population (ch 176). Corrupted by an operation Śiva himself names as a deception, and destroyed for the corruption, in one sentence: *"the great daityas are steadfast in their minds and are devoted to me. Nevertheless, I should kill them, since as a result of deception, they have abandoned excellent dharma."*3 The deception is acknowledged. It is not a defence.
The Dārukavana rishis (ch 283). Provoked deliberately by a disguised god who refuses to identify himself, "deluded by Shiva's maya," and then rebuked by Brahmā — not for being wrong about the conduct, but for failing to recognise who it was.4
Kāma, and Brahmā at the wedding fire. Volume 1's material carries the same shape and the vault already holds it: an agent impaired by a force nobody disclosed, then charged for what the impairment produced.5
Five sets of parties, one common feature. In none of them does anyone say this is going to happen to you, and here is why.
Now the cases where nobody pays, and look at what was said out loud.
Arihat (ch 174). The brief is exhaustive — sixteen thousand verses, against śruti and smṛti, avoiding varṇa and āśrama, in apabhraṃśa, emphasising karma only, deployed to the desert until kali yuga arrives. And attached to it, the indemnity: "Since this is because of my command, no sin will attach to you."1
The operatives, exempted from their own product. "But despite their forms, Narada and the possessor of maya did not themselves become deviants. This was because of the favours of Paramesha."1 They carry the heresy without catching it.
The manufacturers, cleared at ch 182. The shaven-headed ones confess — "we performed this wicked deed because of your wish" — and are told plainly that they will not have a wicked destination, because they served the gods. And in the same breath, "from now on, when kali yuga has entered, men who follow this view will have a wicked destination."6
That last pairing is the rule at its most naked. The people who built the false doctrine, knowingly, on instruction, are indemnified. The people who will one day believe it — who were told nothing, who receive it as inherited truth — are damned. Authorisation is doing all the work, and belief without authorisation is doing none.
The controlled positive case is ch 224, and it is the cleanest experiment the volume runs, because everything is disclosed and the impairment is real.
Śiva has promised Bāṇa protection and cursed Bāṇa's arms. Both commitments are his and they cannot both be kept. Kṛṣṇa arrives to execute the curse and states the bind: "Withdraw from this battle. Otherwise, your curse will fail."
Śiva's solution is to have himself disabled, and he says so:
"You have spoken the truth. I did curse the king of daityas. It is a result of my command that you came here… How can I possibly witness the slicing of Bana's arms? Therefore, follow my command. Use jrimbhanastra to cause yawning in me. Use that opportunity to do what you wish."7
Kṛṣṇa fires it. "Deluded by jrimbhanastra, Girisha yawned." The arms come off while the protector is technically not watching.
Now count the liabilities.
Śiva, who failed to protect a devotee he had guaranteed: nothing. Kṛṣṇa, who deliberately incapacitated a god in the middle of a battle: nothing. Bāṇa, mutilated: nothing — he goes on to repent, be made a gaṇapati, and receive a boon-list.
Every element that generates liability elsewhere is present. An impairment. A god rendered unable to act. A promise broken in substance. And no consequence attaches to anyone, because the whole thing was said out loud, in front of the affected party, before it happened.
Set that against Kāma, and the comparison is close to controlled. Both involve an agent acting at a god's instigation, and an impairment that produces the outcome. The difference is not the severity of the act — Kāma's is trivial by comparison. It is that Kāma's operation ran on an undisclosed condition and this one ran on a stated one.
The most precise ruling in the corpus is four lines long and it is delivered to Agni.
The devas were told exactly what to do: "Act so that the semen certainly falls down on the ground." It fell. The instruction was complete. And then they improvised: "Urged by all the immortals, Agni assumed the form of a dove. Using his beak, he swallowed up all of Shambhu's semen."2
Pārvatī's curse lands on precisely that, and names it: "You have swallowed Shiva's semen. You should not have wished to do that. Nor is it right."2
Then Śiva's ruling, which holds both facts together without collapsing either:
"You have done what should not have been done. You have swallowed my semen. That is the reason, obeying my command, your sin has increased, and the scorching has not been countered."2
Obeying my command — the authorised part is acknowledged. You have done what should not have been done — the unauthorised increment is where the sin sits.
This is why the episode is not, as a first reading of this build recorded it, compliance punished anyway. It is deviation punished exactly, and it supplies the rule's second half: authorisation is a shield, and the shield covers only what was authorised. Step past the brief and you are uncovered for the step, not for the errand.
You are about to do something on someone else's say-so, and it might go wrong.
Here is the whole practice, and it is one question asked before you start rather than after.
"Can you put in writing what you're asking for, and what happens if it goes badly?"
Not because paper protects you — often it doesn't. Because of what asking reveals. Watch what happens to the room. If the answer is a clear brief and a stated allocation of consequence, you are in the ch 224 situation: everything disclosed, everyone's exposure visible, and the risk is a shared and known quantity. If the answer is a change of subject, a reassurance about trust, or let's not make this formal, you are in the Kāma situation, and the reassurance is not a substitute — it is the diagnostic.
Then the second half, which is the one people actually get wrong. Stay inside the brief.
The gods in ch 151–152 did the thing they were told to do. Then, unprompted and in good faith, one of them did one extra thing — a sensible-looking improvement, made mid-task, by somebody who was there and could see what was needed. That single unauthorised increment is what the entire subsequent catastrophe descends from, and the ruling that follows is careful to say so: obeying my command, and also you have done what should not have been done.
So when you find yourself mid-task thinking while I'm here, I'll just also — stop and get it added to the brief. The five minutes that costs is the difference between covered and uncovered, and you will not be able to tell the difference until afterwards.
And the third, for anyone who issues briefs: notice that in this corpus the indemnity is free and the disclosure is what's expensive. Viṣṇu's no sin will attach to you costs him nothing and buys total compliance on an atrocity. What he never does is tell Arihat what the operation is for. If you are handing someone cover instead of information, you have not protected them. You have bought them.
Three obvious alternatives fit some of the cases and fail on others, and ruling them out is what makes the disclosure reading load-bearing rather than decorative.
It is not karma. The volume's karma explanations arrive after verdicts rather than predicting them, and they are offered by interested parties — Śiva tells Tulasī "like everyone else, you have reaped the fruits of your karma" about an act he and Pārvatī are named as having commissioned.8 Karma is the language of the settlement, not the rule that generates it.
It is not power or rank. Brahmā is nearly killed at a wedding fire; Agni is scorched; Kāma is incinerated; the six wives are disgraced — and a manufactured heretic with no standing at all walks away clean. Rank predicts nothing here.
It is not kinship. Jalandhara is "born from my portion" and is killed with a weapon Śiva makes from his own toe. Śaṅkhacūḍa was Kṛṣṇa's own companion and is reduced to ashes. Andhaka is made from Śiva's sweat and is restored. Kinship is present in all three and predicts none of them.
What survives all three tests is disclosure — whether the affected party was told in advance what was being done and by whom.
It is a rule about who is billed, and it should not be stretched into a rule about who is harmed. Two limits, stated plainly.
The victims are not covered by anyone's authorisation. Vrindā and Tulasī are deceived by an operation that is documented at both ends — commissioned by Pārvatī in direct speech, executed by Viṣṇu, glossed by the narrator. The paperwork is complete and it protects the operatives entirely. Neither woman is protected by any of it, and both destroy themselves.8 The indemnity travels along the chain of command and stops at its edge.
And the rule is silent on the god. Śiva is impaired three times in Volume 2 — by Jalandhara's gandharva illusion, by the false Gaurī, and by his own arrangement at ch 224 — and on none of those occasions does anything attach to him for what the impairment allowed. The corpus never applies its own test to him, and the page does not claim it does.
[n≥3 PATTERN] for the rule across the billed cases and the two indemnified ones. What is [n=1 OBSERVED] is the fully controlled instance, ch 224, and the strength of the whole reading rests more on that single well-specified case than on the count.
The strongest evidence is the Arihat/ch 224 pair. One is a pre-issued indemnity in a single sentence covering an atrocity; the other is a fully disclosed, self-requested, publicly negotiated impairment after which nobody is billed.17 Both are direct speech and neither requires inference.
Tension 1 — the makers are cleared and the believers damned.6 Stated as policy, without irony, in the same speech.
Tension 2 — Agni is charged while obeying. Śiva's ruling holds obeying my command and you have done what should not have been done in one sentence and does not rank them.2
Tension 3 — Arundhatī's exemption is never drawn out. The one woman who was told is the one woman not deluded, and the text never treats disclosure as the operative difference, though it prints both halves.2
Tension 4 — the rule protects operatives and not targets, and the corpus shows no awareness that this is a feature of the rule rather than a fact about the world.
Open question. The indemnity is issued by the commissioner. Does anyone in this corpus ever ask for one before acting, rather than receiving it unprompted — and if not, what does that say about who the clause is for?
Against Disabled by Māyā, Then Held Liable: that page establishes the machine; this one supplies the variable. Volume 1 could show that impaired agents get billed but could not say why some are and some aren't, because it had no indemnified case. Volume 2 supplies two — Arihat and ch 224 — and they turn a pattern into a rule with a stated condition.
Against The Commission Pattern: the two findings are complementary and slightly in tension, which is worth naming. That page shows that Śiva's commissions are never quoted when the act falls on a third party. This one shows that a quoted commission is exactly what indemnifies an operative. Put together: the operatives in this volume are working without the document that would protect them, and the narrator supplies the attribution afterwards — which allocates the credit upward while leaving the cover unissued.
Against Kumāra, and the Son Who Left: the ch 151–152 correction lives there and is load-bearing here. The devas obeyed; the injury descends from the unauthorised increment; Pārvatī's curse lands on exactly that. Without that correction this page's second half would be built on a misreading.
Against Tripura, the Hardest Case: the ch 176 sentence is the rule's hardest case, because the deception is acknowledged by the killer in the sentence that orders the killing and still does not function as mitigation. Whatever the rule is protecting, it is not the deceived.
Ram Singh and the Question of Complicity — that page follows a court that needs a name after a disaster. Shivaji walks out of Agra, the cordon explains nothing, and Ram Singh is the peg: present, plausible, and available. The argument is that guilt is not what a frightened court shops for. It shops for an entry that closes the account.
The Purāṇa runs the same market and prices it differently, and the comparison isolates what the price actually depends on.
Kāma is Ram Singh — present, instrumental, available, incinerated. So are the six wives. So is Tripura's population. In each case a cost lands on whoever was standing where the damage happened.
But Arihat is also present, also instrumental, and far more culpable — and no ledger opens on him at all. The difference is not proximity, not rank, not how bad the act was. It is that somebody with authority said, in advance, that this one would not be charged.
What that adds to the historical case is a distinction worth carrying: a scapegoat is not simply the nearest available person. A scapegoat is the nearest available person who has no prior written cover. Agra's court could not have hung it on someone holding a signed instruction, not because the instruction proves innocence but because charging him would require impeaching whoever signed it. Ram Singh was chosen for the absence of a document, and the absence is the thing the historian should be looking for — not the presence of motive.
Moral Agency and Categorical Responsibility — that page opens on the explanatory abyss under blame: at what point in a causal chain does someone become responsible, and is the categorical answer we give an artefact rather than a discovery?
This corpus is an unusually clean specimen for that question, because it removes the two things that normally obscure it. There is no epistemic uncertainty — the text tells you exactly who was impaired, by whom, and whether they knew. And there is no dispute about causation, because māyā is a stated mechanism with a named author.
With those cleared away, what is left determining responsibility is nothing internal to the agent at all. The six wives and Arundhatī have identical agency in the identical scene; the only difference between them is that one was told. Arihat and Kāma have identical instrumentality; the only difference is a sentence spoken beforehand by someone else. In every case the verdict is fixed by an act of speech performed by a third party, before or after the fact.
Which is a sharper version of the page's suspicion than the abstract argument can reach. It is not merely that categorical responsibility has no clean cut-point in the causal chain. It is that in a fully specified system — where causation is known and disclosure is visible — responsibility turns out to track who said what to whom, and when, and not to track the agent's contribution at all. If blame is a bookkeeping operation rather than a discovery, then the interesting question about any verdict is not was it deserved but who was holding the pen.
Sharpest implication. This corpus has a working liability rule and it is not a moral one. Being impaired does not excuse you; being deceived does not excuse you; being the direct cause does not condemn you. What determines whether a cost lands is whether somebody with standing stated in advance that it wouldn't. One sentence from Viṣṇu — since this is because of my command, no sin will attach to you — covers the manufacture of a false scripture that gets three cities burned, while six women who were not told anything discover their own guilt after the fact and pay in full. The rule protects operatives and never protects targets, it is issued free by the party who benefits from compliance, and the only fully controlled case in the volume is one in which the god arranged his own incapacitation and, uniquely, nobody was billed at all.
Generative questions.