You find out that someone you have influence over is doing something you think is wrong. Not slightly wrong. Wrong at the level where you would ordinarily say so.
And you have a choice about your first sentence, and the choice decides the whole conversation.
You can say stop. It is the honest sentence, it is what you actually want, and in most cases it ends the relationship in about four seconds — because the person is not doing this by accident, they have reasons, and the reasons include how they eat.
Or you can say something much smaller. Not stop: not like that. Which feels like a compromise, sounds like a compromise, and gets you a second conversation.
Almost everybody who has changed anybody's behaviour has had to make that choice, and almost everybody finds the second sentence harder to say than the first, because the second one requires you to accept, out loud, that the thing is going to continue this week.
Madhya 24 tells a story in which a sage makes the second choice, and the detail of how he makes it is better than the moral.
Nārada finds Mṛgāri, a hunter, in the forest. The animals around him are not dead. They are half-dead, left where they fell, still moving.
And what Nārada asks for is not what you expect.
"I beg you that from this day on you will kill animals completely and not leave them half-dead."1
Read that as an instruction from a renunciate sage to a man he intends to convert. He is asking the hunter to be better at hunting. No conditions, no prelude, no theology — the first request in the encounter is a request for competence in the very activity the sage objects to.
And he explains, in a sentence that does the whole page's work:
"Your business is killing animals. That is a slight offense on your part, but when you consciously give them unnecessary pain by leaving them half-dead, you incur very great sins."2
The profession is conceded as a slight offence. Not excused — offence is the word — but graded, and graded far below the thing being asked about.
That gradation is the most surprising thing in the chapter, and it is worth resisting the urge to explain it away.
This corpus is not usually gradualist. Its characteristic rule is catastrophic: one offence against a devotee uproots the creeper entire; one association with the wrong company is worse than bars and burning flames.3 It is a tradition of thresholds, not slopes.
And here it produces a clean utilitarian gradient. Killing: bad. Killing badly, so that the animal suffers longer than necessary: much worse. The variable being weighed is the animal's experience, and nothing else in the sentence is doing any work.
Which means Nārada is not correcting the hunter's status, his category, or his interior state — the three things this corpus normally decides everything by. He is correcting an outcome, in the world, measurable by a third party. It is the most checkable moral instruction in the entire book, and it is delivered to the least qualified person in it.
Nothing about the exchange is softened, either. The corpus does not pretend the hunting is fine. It says explicitly that it is an offence, and then declines to lead with it.
The second half of the encounter is where most conversion stories go wrong, and this one does not.
Nārada eventually does ask for the thing he wanted. And the hunter raises the objection anybody in his position raises, which is not theological:
"If I break my bow, how shall I maintain myself?"4
That is the real question. It is the question that ends most attempts to change anybody's livelihood, and it usually gets answered with reassurance, or with a claim about providence, or with nothing.
Nārada's answer is a commitment: "Do not worry. I shall supply all your food every day."5
The bow is broken after the food is guaranteed, not before. The sage takes on an open-ended material obligation to a stranger as the precondition of the stranger's renunciation, and the text records it as an ordinary part of the transaction rather than as a marvel.
Set that beside the corpus's usual account of provision, which is that the Lord maintains his devotees and material anxiety is itself a defect. Here the answer to how shall I eat is not doctrine. It is a person, undertaking to bring food. The story's own logic is that a renunciation demanded of someone who has no other income is not a renunciation; it is a request that they starve.
The coda is what makes this a page rather than an anecdote, because the story does not end where the instruction ends.
Two details.
The hunter cannot prostrate himself until he has swept the ants aside with a cloth, so as not to crush them on the way down.6 Nobody asked for that. The instruction concerned animals he was shooting; it said nothing about insects underfoot. He has generalised the principle past its stated scope, on his own, and the generalisation costs him time and dignity every single time he greets his teacher.
And he asks for less food. "Please do not send so many grains. Only send what is sufficient for two people, no more."7 The subsistence guarantee that made the renunciation possible is now, in his hands, being negotiated downward by the beneficiary.
The convert has ended up gentler and poorer than the instruction required — and both movements came from him.
Which is the argument for the whole method. A person ordered to stop would have complied at exactly the level ordered, if at all. A person who was met where he stood, corrected on one checkable thing, and then supported through the larger change, went past the brief in two directions without being asked.
Parvata Muni's verdict on Nārada is the chapter's own summary: "Certainly you are a touchstone."8 A touchstone does not change what it touches. It reveals what the metal already was.
It is worth listing the absences, because in a corpus this argumentative they are conspicuous.
No threat. Nothing about hell, nothing about the consequences of continuing, nothing about what happens to hunters. The sins are named as a quantity, not as a sentence.
No scriptural argument. Nārada quotes nothing at the hunter, in a book where every other opponent is met with citation, and where the immediately surrounding chapter is sixty-one grammatical readings of a single verse.9
No demand for belief. At no point in the correction is the hunter asked to accept anything about who Kṛṣṇa is. The doctrine arrives afterwards, to somebody whose conduct has already changed.
And no argument about the hunter's category. He is not told that his birth is no obstacle, which is the corpus's standard egalitarian move. The subject never comes up, because nobody raised it.
The encounter is conducted almost entirely in the register the corpus uses least: practical, incremental, and about consequences in the world. That it produces the corpus's most complete conversion is not remarked on anywhere.
Layer A supplies the whole episode — the half-dead animals, the graded offence, the objection about maintenance, the guarantee, the ants and the request for less food. It needs no commentary and gets little.10
Layer B contributes one image to this region that is worth naming precisely because it is so often quoted as though it were the verse: atonement compared to an elephant's bathing — the animal washes thoroughly and then throws dust over itself the moment it comes out of the water.11
That is Bhaktivedanta's purport, not Kṛṣṇadāsa's text. It is vivid, it is memorable, and it argues something the episode does not: that corrective effort without a change of orientation simply undoes itself. The story on the page is the opposite case — a man whose conduct changes first and whose orientation follows it, and who then overshoots.
Both are worth having. Only one of them is in the verse, and they point different ways about whether behaviour can lead.
You are about to tell somebody to stop. Ask what the smallest checkable improvement is that they could make this week without changing anything about who they are. That is Nārada's first sentence, and it is the one that buys the second conversation. It also requires you to say out loud that the thing continues meanwhile, which is the part that feels like defeat and is not.
You want somebody to give up how they earn. Answer the maintenance question before you make the ask, and answer it with a commitment rather than with a reassurance. It will work out is not an answer. I will bring food is. If you cannot say the second one, you are asking them to bear the whole cost of your conviction.
Your correction has a scope and the person has exceeded it. Leave it alone. The overshoot is the sign that the principle transferred rather than the instruction, and instructing them back down converts a principle they own into a rule you issued.
You are choosing between an argument and a request. Note that this corpus's most complete conversion contains no argument at all — no citation, no threat, no theology until afterwards — in a book that argues constantly and converts by argument everywhere else.
The evidence is a complete self-contained narrative inside Madhya 24, with its own dialogue, objection and resolution, and a verdict pronounced on the method by a third party within the story.12
The tension the source doesn't close: the graded offence contradicts the corpus's characteristic rule. Everywhere else the decisive rules are catastrophic and threshold-shaped; here culpability scales smoothly with an outcome in the world. The corpus supplies both and never reconciles them.
A second one. The method that produces the book's most thorough conversion — incremental, practical, non-doctrinal, subsistence guaranteed first — is used exactly once, and the corpus draws no lesson from it. Every other conversion in the book runs on argument or on display.
Open: what Nārada would have said if the hunter had continued hunting well and stopped there. The story's structure requires the first ask to be sincere, and nothing in the range read indicates whether it would have been accepted as sufficient.
Against The Two Formidable Enemies this is the same corpus operating on an opposite theory of change. That rule removes the practitioner from the company of the wrong people, absolutely, on a category. Nārada walks into the forest and sits down with the worst-placed person in the book. Both are in the Madhya-līlā instruction, and the second produces the conversion.
Against The Mad Elephant Offence the contrast is about the shape of culpability itself. That page's rule is one instance, total loss, no gradation. This page's is a smooth gradient with the victim's suffering as the variable. The corpus does not notice that it has two incompatible theories of moral weight, and the gradient one is aimed at outsiders while the threshold one is aimed at members.
And it is the practical form of Good Qualities Follow Bhakti. That doctrine says virtues arrive as a consequence rather than by direct cultivation. The hunter is the worked example — nobody instructed him about ants, and the care for ants is what turned up.
Killing versus Murder: The Psychology of Deliberate Harm — that page's insistence is that contemporary psychology has lost a distinction worth keeping: killing that is soul-driven and independent of personal benefit is a different act from killing driven by rage, fear or desperation. Both take life. The difference lies in the origin of the action, and the distinction is offered as psychological rather than moral.
Both refuse to treat all killing as one act. They put the dividing line in completely different places, and the difference is instructive rather than merely comparative.
Gigerenzer's line runs through the origin. What matters is where the act came from — soul-necessity or ego-pathology. That is a genuine distinction and it has a property worth naming: it is not available to the person being corrected. Nobody can reliably inspect their own motive in the moment, and a criterion located in the origin of an action gives the actor nothing to do differently tomorrow morning.
Nārada's line runs through the victim's experience. Half-dead or dead. That is a worse metaphysics — it says nothing about what the act means, what it comes from, or what it does to the person doing it — and it is an incomparably better instruction, because a hunter holding a bow can act on it before lunch.
Which gives the finding neither reaches alone. A moral distinction and a usable one are different artefacts, and the depth of the first is often what makes the second impossible. Origin-based criteria are the deeper analysis and they route all the work into self-examination, which is exactly where a person already committed to an activity has the least reliable equipment. Outcome-based criteria are shallower and they hand the person a next action.
The story's own evidence is that the shallow one led to the deep one. The hunter changed what he did, then broke the bow, then swept the ants. Nobody sold him the origin distinction; he arrived at it by way of a rule about how animals die.
Sharpest implication. The first sentence of a correction is a choice between what you want and what can be acted on tomorrow, and the second one is harder to say because it concedes that the wrong thing continues meanwhile. This corpus's most complete conversion is built entirely on making that concession — and it is the only place in sixty-two chapters where anybody does.
Generative questions.