A wrestling arena. Two professional fighters, Cāṇūra and Muṣṭika, are hired to kill a boy. They land blows on him — real ones, meant to break him. And the boy is delighted. He's tasting the fight, enjoying the contest, lit up by it.
Second scene, a kitchen. A woman is nursing her son. The milk on the stove starts to boil over. She sets the child down, unfinished, and goes to save the milk. His lip trembles with rage.
Now take a definition of devotion that says: devotion is doing what pleases him.
Scene one passes. The fighters are pleasing him. Scene two fails. The mother isn't.
Both answers are wrong, and everyone in the tradition knows they're wrong. The fighters are assassins. The woman is the standard by which motherly love is measured for the next five hundred years. So the definition has just admitted two murderers and expelled a saint, and it did it by working exactly as written.
Sanskrit logic has terms for these two failures, and Viśvanātha reaches for them before he explains anything else about the verse.1
Ativyāpti — over-extension. Literally over-pervasion. The definition is too wide, so it swallows things that were never meant to be inside it.
Avyāpti — under-extension. Too narrow, so it excludes things that belong.
Two directions, one net. Cast it too wide and you catch the fighters. Pull it too tight and you lose the mother. The two faults are not a list of problems; they're the two walls a definition has to fit between.
And notice what these terms are not. They aren't moral categories. Nobody is saying the definition is unkind or impious. They're saying it has the wrong extension — it picks out the wrong set of things. That's a technical complaint, and it gets a technical fix.
Take the demons first, because they're the easier of the two.
Cāṇūra and Muṣṭika strike Kṛṣṇa in the ring, and the text is clear that he enjoys it — he begins to taste vīra-rasa, the mood of heroism, in their company. The book cites a line to explain why this isn't strange: for ordinary people a fierce fight with an enemy is misery, but for great heroes it's a pleasure.2
So the pleasure is real. Which means the definition keyed to pleasure has to admit them.
And admitting them is absurd, because their attitude is nothing but hostility. They were paid to kill him. The definition would have to say that being struck by a hired killer counts as an act of devotion, on the grounds that the victim enjoyed the exercise.
That's ativyāpti. The net is too wide.
The second case is harder, and I think it's the more instructive of the two.
Yaśodā sets Kṛṣṇa down mid-feed and goes to the stove. He is displeased — the text gives his trembling reddened lip. Under a pleasure-keyed definition, she has just done something that is not devotion.
But look at why she got up. The milk was for him. She left him in order to save the thing that would nourish him. Her action was displeasing to him in the moment and entirely oriented to him underneath.
And then the book says something I didn't expect. Devotees, it observes, naturally display greater love for the things favourable to serving him than for him directly.3 So her getting up isn't a lapse she gets forgiven for. It's a symptom of the intensity of the thing being measured. The more devoted she is, the more likely she is to put him down for the milk.
Which means a pleasure-keyed definition doesn't just get her case wrong. It gets it wrong in proportion to how devoted she is. The better the devotee, the worse the definition performs.
That's avyāpti. And it's the one that should worry you.
Viśvanātha's fix is small and it changes everything. The verse doesn't say devotion is what pleases him. It says ānukūlyena — favourably disposed.
And the real meaning of ānukūlya, he insists, is not producing pleasure. It is the complete absence of any attitude that is unfavourable or hostile.4
Run the two cases again.
The fighters: permanently, constitutionally hostile. They fail on disposition, whatever the effect of their blows. Over-extension closed.
Yaśodā: not a trace of anything unfavourable, ever — she is permeated with attentiveness to his welfare, which is precisely why she went to the stove. She passes on disposition, whatever the effect on his lip. Under-extension closed.
One substitution — effect swapped for disposition — and both walls hold at once. That's the move, and it's the reason the definition survives.
Your father is ill and you cancel the visit because you're finishing the work that pays for his care.
He's hurt. You hear it in his voice.
On the disposition standard you are doing the Yaśodā thing exactly — leaving the person to save the milk. And you cannot prove it. There is no external test that separates you from someone who simply preferred to work.
The repair that saved the definition also made it unauditable from outside. You're now standing inside that, at nine at night, with the phone in your hand.
Notice, too, which standard you reached for. On yourself, disposition. On the colleague who cancelled on you last month, effect.
With only the demons, you could patch the definition cheaply — and not performed with hostile intent. Done.
Yaśodā is what makes that insufficient.
Her case can't be repaired by a clause about bad intent, because there is no bad intent anywhere in it. Her intent is impeccable and the effect is displeasure. So the fix can't be a motive filter bolted onto an effect-based criterion. The criterion has to move off effects entirely.
That's why the cases come together, and always in that order. The demons show the net is too wide. Yaśodā shows that narrowing it won't help, because the axis is wrong. One case demands a patch; the pair demands a redesign.
The milk does a second thing. It establishes that you can displease God and be doing devotion — that the surface of the relationship and its substance can come apart, and the substance wins. That concession is load-bearing later, when the tradition has to count a devotee's anger, jealousy and refusal as devotional states rather than failures.
Viśvanātha doesn't test the definition against a typical case. He hunts the two extremes — the worst thing that satisfies it, the best thing it rejects — and squeezes from both sides at once.
That's a vise, not a checklist. And a vise has a property a checklist doesn't: survive both jaws and the cases in between are already covered. You never have to enumerate them.
Which is why he spends his effort on assassins and boiling milk instead of describing ordinary worship.
Strongest support. The two faults are named, worked in opposite directions on two specific narrative cases, and closed by a single lexical choice already present in the verse. Nothing is added; the repair is a reading.
The tension the book does not close. The repair moves the criterion from observable effect to unobservable disposition. That is what makes it correct, and it is also what makes it untestable. Yaśodā passes because she has no trace of an unfavourable attitude — and there is no procedure, anywhere in the book, for establishing that about anyone. The tradition's answer is that the verdict on Yaśodā comes from scripture, which is fine for Yaśodā and no help at all for you.
A second tension. The claim that devotees love the means of service more than the served — offered here to rescue Yaśodā — is a substantial doctrine smuggled in as a repair. Push it and it justifies almost any neglect of the beloved in favour of arrangements made on the beloved's behalf. The book states it once, uses it, and never bounds it.
Open question. Are these two faults being used as genuine logical tools, or as a frame for a verdict already fixed by narrative authority? The demons were always going to fail and Yaśodā was always going to pass. The logic arrives after the answers are known — which is not disqualifying, but it means the method is being used to justify a boundary rather than to find one.
Read the five-rasa material by itself and Yaśodā arrives as a finished icon — parental love, the highest of the graded moods below the amorous, illustrated by a woman who ties God to a grinding mortar. She's an exemplar. Come at her through Text 1 and she's something else entirely: a problem case, wheeled in because she breaks a definition. The affection is the same; the job is not. Here she's evidence, and specifically she's the evidence that pleasure can't be the criterion.
That reframing runs the other way too. Once you've met Yaśodā as a counter-example, the demons stop looking like villains in a story and start looking like a test input — the tradition's own adversarial case, chosen because it's the most repellent thing that satisfies the rule as written.
Over- and under-extension is the false-positive/false-negative trade-off, and both fields discover the same thing: you cannot minimise one without paying in the other unless you change what you're measuring. Set this against Deception Rating Scale and Behavior Analysis Process. That page's entire difficulty is that a behavioural indicator of lying will either flag honest people or miss liars, and tightening the threshold to fix one failure worsens the other. Same vise, different subject. Ativyāpti is the false positive — the fighters flagged as devotees. Avyāpti is the false negative — Yaśodā missed.
Here's what the pair produces that neither gives alone. Both escape the trade-off the same way — by abandoning the surface signal for an underlying variable. Viśvanātha swaps effect-on-Kṛṣṇa for disposition-toward-Kṛṣṇa; deception work moves from the isolated tell to baseline-relative clusters and intent. And both pay the identical price: the new criterion gets the hard cases right and can no longer be checked from outside.
Which suggests something uncomfortable and general. When a classification system finally handles its edge cases, that is often the moment it stops being verifiable — because what was tripping it up was its own commitment to observable evidence. Both traditions then reach for the same patch, an authorised interpreter, and neither counts it as a cost of the fix.
Sharpest implication. Test a definition at its two extremes, not in its middle. The worst thing that satisfies it and the best thing it rejects will together tell you whether your criterion is even pointed at the right variable — and if it isn't, no amount of tightening or loosening will help, because both jaws are reporting the same underlying mistake.
Generative questions.
ānukūlya repair in its full settingvīra-rasa, the mood the fighters provoke, sits in the finished scheme