Somebody is doing something to themselves and you can see it.
And you already know what it means. You knew before you finished looking, because behaviour of that kind arrives pre-interpreted — it is an expression of something, an outward form of an inward pain, a person putting on the outside what they cannot say on the inside. That reading is so automatic that it does not feel like a reading.
Antya-līlā 19 supplies the scene and then does something almost no source does. It supplies the man's own account of what he was doing — and the account is not an expression of anything.1
All night, in the Gambhīrā, he rubs his face against the walls.
Blood from his mouth, his nose and his cheeks. And the sound he is making, recorded as syllables: goṅ-goṅ.2
That is what the devotees find, and it is the kind of thing that ends an account. There is nothing obviously left to ask; the scene is complete and terrible and interprets itself.
And then he is asked, and he answers.
*"I was in such anxiety that I could not stay in the room. I wanted to go out… Unable to find the door, I kept hitting the four walls with My face."*3
Read what that sentence actually reports.
He was trying to leave.
The face against the wall is not a wound being inflicted; it is a method of navigation. He is looking for a door he cannot find, in a room he knows, using the only part of him that is still reporting anything, and the four walls are what he keeps finding instead.
The blood is incidental. It is what happens when you do that for a long time. It is not the point of the activity and it was not the object of any intention.
So the scene the household walked into was not self-harm as expression. It was self-harm as a failed search for an exit — an instrumental act, with a goal, being performed continuously, by somebody whose environment had stopped being navigable.
The general problem is worth stating carefully, because it is the most portable thing on this page.
Behaviour underdetermines whether there is a goal.
Watch somebody press their face into a wall repeatedly and you have all the observable facts. What you do not have — and cannot get by watching harder — is whether the wall is the object of the behaviour or an obstacle to it, and those are opposite situations that look identical from outside.
An observer defaults to the expressive reading, and there are reasons. Expressive readings require no additional information. They fit what the observer already believes about distress. And crucially, the instrumental reading requires you to posit a goal you cannot see — which feels like invention, and is exactly what the situation demands.
The corpus's community, on the evidence of this passage, did not work it out. They found him, they stopped him, and it took a question to produce the account. Nothing in the scene as reported suggested a door.
Which generalises directly. Any behaviour aimed at something the observer cannot perceive will be read as expressive, because expression is the residual category — what behaviour gets classified as when no goal is available. And the people best placed to have goals invisible to observers are precisely the people whose perception has departed from everybody else's.
That is not a marginal case. It is the standing condition of anybody in an altered state, a delirium, a dementia, or a language nobody in the room speaks.
And there is a second asymmetry stacked on the first, which is why the error is so stable.
The expressive reading cannot be disconfirmed by watching. If he is expressing anguish, continued face-against-wall is more of the same and confirms. If he is searching for a door, continued face-against-wall is also what you would see, because he has not found it. Every additional minute of observation is equally consistent with both, so an observer who watches longer becomes more confident without becoming better informed.
Which is the specific trap. Duration feels like evidence. A behaviour that goes on all night reads as overwhelming, and the reading strengthens with every hour — and on the instrumental account the hours mean only that the room was hard.
The information that separates the two was never in the room at all. It was in a sentence, and the sentence had to be asked for.
There is a transmission fact here that is worth pulling out, because it explains why this page can exist.
Somebody asked him, and somebody wrote down what he said.
That is not automatic. The overwhelming majority of accounts of somebody in extremity are entirely observational — what was seen, what was done, what was concluded — because the person is not in a position to be interviewed and because the observers already have their reading.
This household asked. And their secretary recorded the answer verbatim enough that four centuries later it is possible to see that the community's interpretation and the man's own account do not match.4
Which is the corpus's most valuable habit and it is entirely accidental to its purpose. Nothing in the theology requires the first-person account; the scene works better without it. The account survives because the practice was to write down what he said, not to write down what it meant — and a transcript preserves the material that contradicts the transcriber.
Compare the goṅ-goṅ. The same syllables are reported, independently, by a fisherman who netted him out of the sea in another chapter and had no idea who he was.5 Neither report is doing anything for the theology. Both survive because somebody was writing down sounds.
The response is recorded and it is the corpus's most uncomfortable pattern, arriving here in its most literal form.
They put a man in the room.
Śaṅkara Paṇḍita lies at his feet; the Lord rests his legs on him; Śaṅkara wakes repeatedly through the night to resume massaging. And the stated effect is explicit: *"Out of fear of Śaṅkara, Śrī Caitanya Mahāprabhu could neither leave His room nor rub His lotuslike face against the walls."*6
Note the two clauses in that sentence and the order they come in. Neither leave the room nor injure his face. Leaving is named first.
The containment addresses the goal, not the injury — and it does so without anybody having identified the goal, which is what makes it work. A man who wants to get out and cannot find the door is stopped by a person lying across the exit, whether or not that person knows there was an exit being sought.
And the corpus supplies the detail that makes the whole thing unbearable rather than sinister: Caitanya gets up and covers the sleeping Śaṅkara with his own quilt.7
Everybody in the room is doing the most loving thing available to them, and the arrangement is still a man being prevented from leaving by fear of a friend.
And fear is the corpus's own word for the mechanism, which is worth not softening. Not persuasion, not request, not agreement. The text says the restraint operated because he did not want to disturb Śaṅkara — which is to say the containment ran on his consideration for the person containing him, and was therefore strongest precisely because he was good.
That is the most efficient restraint there is and the least visible as restraint. A guard can be resented and a friend lying across your feet cannot, and the corpus records the arrangement approvingly, names its mechanism accurately, and never asks what it would take for him to have got past it.
Layer A supplies the night, the walls, the blood, the syllables, the first-person explanation, Śaṅkara's installation and the quilt.8
Layer B expands the episode into the intensity of separation and the devotees' service to the Lord in that condition.9
Worth marking what the commentary does with the explanation, which is nothing. Layer B has no use for the door. On its reading the behaviour is the transport of separation manifesting physically, and the man's own account of trying to get out of a room is a detail inside a scene about something else.
Layer A gives the door a sentence and gives it the last word in the exchange. A reader working from the purports meets a man overwhelmed by divine love. A reader working from the verse meets a man who could not find the door, and only the second is the account of the person it happened to.
You are watching somebody harm themselves. Ask what they were trying to do before you conclude what they were expressing. The expressive reading is the residual category — what behaviour gets filed as when no goal is visible — and the people most likely to have invisible goals are the ones whose perception has departed from yours.
You are caring for somebody who cannot explain themselves. Write down what they say and not what you think it means. The interpretation you supply will be indistinguishable from the content within one retelling; the transcript is the only version that can later be found to disagree with you.
You are containing somebody for their safety. Notice whether you have identified what they were trying to do. Containment works without that knowledge — a body across the door stops an exit-seeker and an expression alike — and working without it means you will never learn whether the thing they wanted was reachable another way.
You have concluded that a behaviour means something. Check whether the meaning required no additional information. Readings that cost nothing to arrive at are the ones that arrive fastest and get tested least.
The evidence is a scene reported observationally, a first-person explanation given in direct speech in the same passage, the containment arrangement with its stated effect, and one independently corroborated detail.10
The tension the source doesn't close: the community's reading and the man's own account do not match, and the corpus records both without noticing. The scene is narrated as ecstatic transport and explained by its subject as a failure to find a door.
A second one. The stated purpose of Śaṅkara's installation is that he could neither leave nor injure himself — and the corpus treats those as one goal. They are different goals with different justifications, and the first is confinement while the second is care.
Open: whether anybody in that household ever repeated his explanation back to anybody. It is recorded once, in his own words, and nothing downstream in the corpus uses it — including the chapters that describe the same behaviour again.
This page and Care as Containment are the same night from the household's side and the subject's. That page holds the seven-instance pattern — restraint, interception, deception, throttling, all called service; this one has the one occasion where the person being managed says what he was doing, and it does not match what anybody thought.
Against Vṛndāvana Overwrites Purī it is the same complaint in a different register. There he asks who brought him back from Govardhana; here he explains that he was trying to get out of a room — and both are statements about wanting to be somewhere else, made by somebody the corpus otherwise reports as transported.
And it belongs beside The Fisherman and the Ghost, which independently reports the same syllables. Goṅ-goṅ, recorded twice, by two parties, in two chapters, neither of them doing anything with it — and it is the closest thing to instrumentation this corpus contains.
Learned Helplessness and Captivity — that page's mechanism is Seligman's: dogs given shocks they could not predict or escape later failed to escape shocks they could. The finding is not that they were broken but that they had learned accurately — in that environment, action and outcome were uncoupled — and the inference then generalised past the conditions that produced it.
The Gambhīrā night is the exact inverse, and the pairing gets at something neither page contains.
Learned helplessness is a case where effort stopped and the observer is tempted to read the stillness as damage. The corrective is that the stillness is a rational inference, and the correct response is to demonstrate that the environment has changed.
Here effort never stopped. He searched all night. The environment was unnavigable and he kept acting anyway, and the observers read the acting as expression rather than as search.
Which gives the finding neither page reaches. Both errors are the same error, made in opposite directions: the observer reads the behaviour and infers the internal state, and behaviour is the wrong evidence for that inference in both cases. Stillness gets read as brokenness when it is accurate learning; action gets read as expression when it is instrumental. The variable that would settle either — what the person believes is available to them — is never on the outside of a person.
And this corpus obtained it the only way it can be obtained. Somebody asked. Which is not a technique, an assessment or a framework; it is the single procedural step that both failure modes require and that neither literature emphasises, because it is too obvious to write down and too easy to skip in a room where everybody already knows what they are looking at.
Sharpest implication. Expression is the residual category — what a behaviour gets filed as when the observer cannot see a goal — and the people most likely to have goals you cannot see are the ones whose perception has already left yours. This corpus's household read a night of self-injury as transport, and the man said he was looking for the door. The only reason anybody knows is that somebody asked and somebody wrote down the answer instead of its meaning.
Generative questions.