Somebody changes their mind completely and joins your side, and it is a real conversion, and then they say something that isn't quite right.
Not wrong exactly. A small proprietary variation — a phrasing you don't use, an emphasis you'd drop. And you correct them, gently, and they accept the correction, and then a moment later they say the thing again, in their own words, because it is what they actually think.
And you have to decide what to do about that, and there isn't a good answer, because the alternative to letting it stand is telling somebody who has just given up their whole position that they have not yet given up enough.
Madhya 6 has that, over a single word in a verse.
Sārvabhauma has converted. He is reciting the Bhāgavatam and comes to a verse containing the word mukti-pada — a name of the Lord meaning, in the plain reading, the shelter of liberation, the one from whom liberation comes.
And he no longer likes it. Liberation is now, in his new position, a poor thing to be naming God after. So he proposes bhakti-pada instead — the shelter of devotion — and the emendation is an improvement by every standard his new tradition holds.1
Caitanya forbids it.
The word stands. Mukti-pada is what the verse says and what the verse says is what it says.2
And then the last beat, which is why this is a page: Sārvabhauma does not agree. He accepts that the word cannot be altered and states that he still prefers his own reading — and the corpus records this, approvingly, without softening it.3
The refusal follows directly from the chapter's own hermeneutic rule, and it is worth seeing that it costs the corpus something.
The rule: take the plain sense; interpretation covers the sun with a cloud.4 A rule of that kind is only a rule if it binds you when it hurts, and here it hurts. The corpus dislikes liberation intensely — elsewhere it classes the desire for it as a weed growing beside the creeper of devotion, and has its devotees decline all five varieties when offered.5
So the text has retained, in its central scripture, a divine epithet built out of the word for a thing it regards as a weed — because the epithet is there.
That is the rule doing its job at its most expensive, and it is the strongest evidence in the corpus that the direct-meaning principle is a genuine commitment rather than only a weapon.
Worth holding that against the same chapter's eighteen readings of a single verse, which is the same rule at its least convincing.6 The corpus is neither consistent nor cynical; it is a document with a principle it sometimes honours at cost and sometimes suspends for display.
The general form is worth naming, because the corpus does not.
An emendation made by someone who now holds a position is an improvement by the standard of that position, and it is unavailable to anybody who does not hold it. Sārvabhauma's bhakti-pada is better only if you already agree that devotion outranks liberation. A reader who did not would receive a text that had been quietly adjusted to say so.
Forbidding it keeps the scripture capable of disagreeing with the tradition that reads it. A text that has been corrected to match the current position can no longer be evidence for the position; it has become a record of it.
That is the actual value of the refusal, and the corpus's own history is the demonstration: the tradition kept a word it disliked, and four centuries later that word is still in the verse, still slightly awkward, still available to anyone who wants to argue that the corpus's attitude to liberation is narrower than the scripture's.
The part of this episode that is genuinely unusual is not the ruling. It is what the corpus does with the disagreement.
Sārvabhauma has just been corrected by the figure the book exists to establish as God, on a point of scriptural integrity, and he says he prefers his own reading anyway. And the corpus writes it down.
Compare what happens elsewhere to people who hold a partial or private view. The author's own brother holds full faith in one figure and partial faith in another and is told he is worse than an atheist — and falls down.7 The branches that did not accept the original trunk withered.8 The corpus is not, on the whole, tolerant of unresolved positions.
Here it is. A converted man keeps his own opinion, in public, on the record, and the text does not report it as a failing.
Two readings, both available.
Generous: the corpus distinguishes between the text, which is not negotiable, and a reader's view, which is theirs. Compliance with the first is required and agreement is not, and that is a genuinely liberal settlement.
Deflationary: the disagreement is harmless because it concerns a preference rather than a practice, and the corpus's tolerance is calibrated to consequence rather than to principle. Sārvabhauma's private view about a word affects nothing; the brother's view about Nityānanda affected who got honoured.
The text supports both and this page does not choose. What is worth recording is that the corpus's most tolerant moment and its least tolerant moment are separated by a difference in stakes, and the corpus never states a principle that would cover both.
Layer A gives the proposal, the refusal, and the retained disagreement, briefly.
Layer B is firm on the ruling and develops the general prohibition on altering scripture, which is doctrinally standard and correct.9
What the commentary passes over is the last beat. The retained disagreement is the interesting part of the episode and it is not what the passage is used for, in an edition that has ample reason to prefer the ruling to the scene.
You have converted somebody and they are still saying it slightly wrong. Decide whether the variation touches practice or only phrasing. If it is phrasing, leaving it alone costs nothing and buys you a person who is actually convinced rather than one who has learned the words.
You want to fix a document to match what everybody now agrees. Stop. A record corrected to match current opinion is no longer evidence of anything, and its only remaining function was to be able to contradict you.
You are enforcing a principle and it has just become expensive. That is the only moment the enforcement means anything; every prior application was free.
Somebody in your organisation complies and says openly that they still disagree. That is the best available outcome and it is frequently treated as the worst, because silent compliance looks like agreement and is not.
The evidence is a short exchange with a stated proposal, a stated refusal and an explicitly recorded residual disagreement.10 Compact and unambiguous.
The tension the source doesn't close: the corpus defends the word and disparages the thing. Mukti-pada is preserved in scripture and the desire for liberation is a weed. Both are canonical and the corpus never places them together.
A second one. The same chapter forbids a one-word emendation on plain-sense grounds and produces eighteen readings of a different verse. The rule binds at the point of textual alteration and not at the point of interpretive multiplication, and no principle distinguishing them is offered.
Open: whether Sārvabhauma's tolerated disagreement is a settled policy or a courtesy to a new convert. Nothing in the range read generalises it, and the corpus's treatment of other partial positions runs the other way.
This page is Direct-Meaning Hermeneutics tested at cost, and it is the strongest evidence that page can call on. A rule that produces only convenient results is not a rule; here it produces an inconvenient one and is obeyed.
Against Refusing the Five Liberations the pairing is the corpus's own collision. That page has exalted devotees decline all five liberations when personally offered; this one has the corpus refuse to remove the word liberation from a name of God. The thing is a weed and its name is protected, and the extraction record carries the pair as collision material.
And it stands against Half-a-Hen Logic as the corpus's opposite disposition. That argument closes the middle position: better an atheist than a hypocrite, no partial acceptance. Here a partial acceptance is recorded without rebuke. The corpus's tolerance and its intolerance are both absolute and are applied to different things, and the sorting variable is not stated anywhere.
Never Appear Too Perfect — Greene's judgment is that flawlessness generates envy and silent enemies, that displaying a defect deflects it, and that only gods and the dead can seem perfect with impunity.
A scripture with an awkward word left in it is the textual version of that, and the comparison shows what the corpus gained by the refusal — and why the gain is not the strategic one.
Greene's defect is deployed: chosen, harmless, and displayed in order to produce an effect on observers. The value is entirely in being seen.
Mukti-pada is retained, which is a different operation. Nobody chose it, it is not harmless — it names God after something the tradition calls a weed — and no observer was meant to be reassured by it. It is there because removing it was forbidden.
And the effect is the one Greene's technique aims at and cannot reliably produce, precisely because his version is a technique. A text that has kept a word against its own preferences is checkable: you can see the friction, you can see that it would have been trivially easy to smooth, and you can see that it wasn't. That is evidence of a constraint operating on the editors, and it is the only kind of evidence about editorial integrity that a reader can actually get.
Which gives the general rule neither page states: a deliberately displayed flaw signals nothing durable, because it is available to anyone; an inconvenient retention signals a constraint, because it is only available to somebody who was constrained. The test is not whether the imperfection is visible. It is whether removing it would have been easy and in the author's interest — and whether it is still there.
Four centuries later, it is.
Sharpest implication. The integrity of a record is only ever demonstrated at the moments it costs something to maintain, and every one of those moments looks, from inside, like an opportunity to make a small obvious improvement. This corpus kept a word it disliked in a name of God, and that single refusal is worth more as evidence about its editorial practice than any number of statements about scriptural authority.
Generative questions.