There is a class of human experience that leaves no record, and it is not a small class.
Whatever happens in deep sleep. Whatever happens under anaesthesia. Whatever happens at the bottom of a seizure, a blackout, a coma. The person was there and nothing came back, because coming back requires somebody who can speak, and the states in question are defined by there being nobody available to do that.
So everything anybody claims to know about the far end of interior experience arrived through a gap: a stretch where the person was still mostly gone and had just enough surface left to say something out loud.
Antya-līlā 18 names that gap, defines it, and — without appearing to notice — identifies it as the channel through which the entire book reached its author.1
The corpus states the doctrine directly:
"The Lord remains in one of three different states of consciousness at all times: internal, external, and half-external… deeply absorbed in internal consciousness but nevertheless exhibiting some external consciousness."2
Three positions, and the third is defined as a mixture rather than a midpoint.
Internal — absorbed, unavailable, producing nothing an observer can use. External — ordinary, present, conversational, and by definition not in the state anybody wants reported. Ardha-bāhya, half-external — absorbed and speaking.
That third term is not a curiosity. It is the only condition in which the interior of the state and the capacity to report it exist in the same person at the same moment. In the other two, one of the two is missing.
Now put it together with the corpus's own account of how it was made, which it supplies openly and more than once.3
Svarūpa Dāmodara kept a notebook. He was the secretary, present through the Purī years, and he recorded what Caitanya said. Raghunātha dāsa memorised it. And Kṛṣṇadāsa Kavirāja wrote the book from Raghunātha's recitation, eighty years after the events, in Vṛndāvana, under the Gosvāmīs' supervision.
Every link in that chain is stated by the author himself. What the author does not state is the link before the first one.
Svarūpa Dāmodara could only write down what was said aloud. The internal state produced nothing. The external state was not the state. Therefore every one of the visions, the addresses to Rādhā, the analyses of Kṛṣṇa's pastimes, the arraignments of Providence — all of it came out of ardha-bāhya speech, and nothing that stayed inside is in the book, and there is no way to know how much that was.
The corpus defines the channel in one chapter and describes the chain of custody in another and never places them side by side. Set together, they say that the Caitanya-caritāmṛta is a transcript of a partially absent man, and the tradition's own vocabulary is what establishes it.
And the middle link does something that is usually misread, so it is worth being exact.
Memorisation is not a lossy stage. It is a stabilising one. The instinct is to treat and then a man memorised it as the weakest point in the chain — one more opportunity for drift, on the model of a rumour passed round a room. That model is wrong for this case. A rumour degrades because nobody is trying to hold it. A memorised corpus is held deliberately, rehearsed, and checked against other holders, and what it does to its contents is fix them.
Which relocates the risk rather than removing it. Whatever entered Raghunātha's memory stopped changing at that moment; whatever did not enter it was gone at that moment. The chain's real vulnerability is not distortion in transit but selection at the point of memorisation — and selection is invisible afterwards in a way that distortion is not, because a dropped item leaves no trace of having been dropped.
A stabilised record is a record whose gaps are permanent. And the corpus gives no account at all of how Svarūpa Dāmodara decided what was worth writing down on any given night — which is the one link in the chain it names without describing.
The consequence is not that the record is unreliable. It is something more specific and more interesting.
A channel with a bandwidth limit does not degrade its contents evenly. It selects.
What passes through ardha-bāhya is whatever is speakable while mostly absent — which favours certain things sharply. Formulaic material passes: verses already known, names, the bhramara-gīta and the standard laments, all of them pre-formed and available without composition. Emotionally saturated material passes, because affect drives vocalisation. And repetition passes, which is why the same syllables recur.4
What does not pass is anything requiring sustained construction from a man who is not, at that moment, constructing much.
Which produces a testable prediction about the record, and the record bears it out. The madness chapters are made of quotation, address and lament — recited verses, speeches to absent persons, catalogued speech-types — rather than of new argument. The extended reasoning in this corpus all sits in the chapters where he is external: the Sārvabhauma debate, the Rāmānanda dialogue, the Sanātana instruction.5
The distribution of genres across the book tracks the state he was in, and the tradition reads that distribution as two kinds of teaching. It is at least as well explained as one channel with a bandwidth.
And the one apparent exception is worth naming rather than smoothing. Mid-vision at Antya 18 he delivers literary criticism on the vision itself, naming atiśayokti and virodhābhāsa as rhetorical figures — which is analysis, not lament, produced in the state.6 Either the bandwidth is wider than the prediction allows, or that material is exactly the kind of pre-formed technical vocabulary a trained man carries ready-made. The page cannot settle it and should not pretend to.
The finding is not deflationary in the direction people expect, and both readings have to pay.
The devotional reading pays this: the tradition's most exalted content is, by its own doctrine, the output of a state its own doctrine defines as partial. Nothing from internal consciousness is in the book. The summit is the one thing the record structurally cannot contain, and every description of it was made by somebody on the way back.
The sceptical reading pays this: the corpus supplies the limitation itself. It did not have to define the three states, and having defined them it did not have to specify that the third is where the speech happens. A text concealing its transmission problem does not name its own channel, and this one names it in a chapter about being pulled out of the sea.
Which leaves both parties in the same place, holding a record whose limits are stated on its face. That is a better epistemic position than either usually gets, and it is the reason this corpus is worth reading closely rather than either accepting or dismissing.
There is one more piece and it belongs here even though its page belongs to Cluster M.
Kṛṣṇadāsa stops the narrative, at the point where the bodily transformations are least believable, and states his own evidentiary position out loud: that no one has witnessed such changes elsewhere, that they are not described in the śāstras, that people in general do not believe them, that Raghunātha dāsa lived continuously with Caitanya and he is recording what he heard from him — and that he believes it totally regardless.7
That is a man doing his own source criticism and then declining its verdict, in public, in the text.
He names the chain, names the single eyewitness, names the absence of scriptural support, and affirms anyway. Put that beside the three-states doctrine and the picture is complete: the corpus knows what kind of evidence it has, states it accurately, and asserts the content on faith rather than on the evidence.
Which is an unusual thing for a partisan document to do, and it is why the ardha-bāhya definition should be read as candour rather than as an oversight the reader has caught.
Layer A supplies the three-state doctrine and its definition, the chain of custody, and the author's statement of his own evidentiary position.8
Layer B expands ardha-bāhya into instruction on levels of absorption available to a practitioner and on the Lord's inconceivable nature.9
Worth marking what the commentary does not do with it. Layer B treats the three states as a description of the Lord and not as a fact about the record. That is a coherent reading and it forecloses the question this page is about, because on Layer B's account the transmission is guaranteed at the far end regardless of the channel.
The transmission problem is only visible if you read the doctrine as a fact about speech — which is what Layer A's sentence, taken plainly, actually says.
You are relying on somebody's report of an interior state. Ask what condition they were in when they said it. Reports from inside extreme states are always made on the way out, and the way out is a different state from the one being reported on.
You are assessing a body of testimony. Look for the channel, and then look at what a channel of that width would pass. Formula, affect and repetition survive narrow bandwidth; sustained novel construction does not. The genre distribution across a corpus is evidence about the conditions of its production.
You are keeping a record of somebody who cannot keep their own. Write down what was said, not what you concluded. Svarūpa Dāmodara's notebook survived eighty years and two intermediaries because it was a transcript; every interpretation added at the time would have been indistinguishable from the content by the time it reached the author.
You have a source that states its own limitations. Rate it up, not down. A document that names its single eyewitness, admits the absence of corroboration, and asserts anyway has told you exactly what you are holding — which is more than almost any source does, and the assertion is then yours to weigh rather than to detect.
The evidence is one explicit three-state definition, the corpus's own repeated account of its chain of custody, and the author's stated evidentiary position.10
The tension the source doesn't close: the tradition's highest state is defined as the one that produces no speech, and the entire record of it is speech. Both facts are stated in the corpus and never set against each other.
A second one. The commentary reads the three states as a fact about the Lord; the verse reads as a fact about consciousness and therefore about the record. Only one of those readings has consequences for the reader's evidence, and it is the verse's.
Open: how much never came back. The channel's existence is stated; its width is not, and there is no way to estimate what stayed inside — which is a limit on the corpus that the corpus itself is the only reason we know about.
This page and Divya-unmāda are the state and its aperture. That page asks what the category does to everybody in the room; this one asks how anything about the state got out of it, and the answer constrains everything the other page is built on.
Against Mādana and Mohana it is the taxonomy's transmission problem. That page has the vocabulary for the state; this one observes that the vocabulary was built from reports made by somebody half out of it — and the corpus supplies both without connecting them.
And it belongs beside He Refuses the Identification, which is the corpus's other chain-of-custody problem. There the evidential weight falls on third-party reports of involuntary phenomena, all sourced from inside the community that already held the claim. Here it falls on first-party speech from inside a state defined as partial. Two different custody problems, one candid text, and no adjudication of either.
Hypnotic Trance Fundamentals — that page's mechanism is that trance is not unconsciousness but the opposite of distraction: the ordinary experience of being so absorbed that somebody has to say your name twice, and the small real jolt of coming back. Absorption is a capacity the brain already has and uses constantly, and depth of absorption is continuous rather than binary.
Ardha-bāhya is that continuum named from the far end by a tradition with a reason to map it, and the pairing produces something neither page has.
Trance research studies absorption as a state to be entered and exited, and its evidence is overwhelmingly the subject's report afterwards, from the ordinary state, reconstructed.
The corpus's interest is the opposite one. It has no use for the retrospective account; it wants what was said during, and it has built a term for exactly the band in which saying-during is possible.
Which gives the finding neither page reaches. The reportable band and the deep band are different bands, and every account of absorption we have — devotional or clinical — is drawn from the shallow edge of it. Retrospective reports are reconstructions made by a different state; concurrent reports come only from the partial one. There is no method, in any tradition, that gets testimony out of full absorption, and the corpus is unusual only in having named the constraint rather than in being subject to it.
Sharpest implication. Every account of deep interior states, in any tradition, is drawn from the shallow edge — either reconstructed afterwards by a different state, or spoken during a partial one. This corpus names that constraint in its own vocabulary and then supplies its entire content through it, and a channel of that width selects for formula, affect and repetition over sustained construction. The genre distribution across the book tracks the state he was in.
Generative questions.