You see blue. Pull that apart and there are obviously two things in it: the blue, and the seeing of it. Yogācāra gives them names — as a seeming object it's the perceived or image aspect, the nimittabhāga; as a seeming subject it's the perceiving aspect, the darśanabhāga.1
What makes this worth reading is that neither half is asserted. Both are established by showing what breaks without them.
Drop the image half and you get an absurdity: a mind with no internal image would be "incapable of perceiving their own object of perception, or else they would… be capable of perceiving indiscriminately all objects."1 With nothing inside standing for the thing, there's nothing to make your perception about your object rather than about everything at once.
Drop the perceiving half and you get another one: "like space, they would not be aware of anything at all. Or else we may say that space itself takes an object and is aware of it."1 A room contains a chair and doesn't perceive it, and the difference between you and the room is the half you just deleted.
Nobody is asked to sit down and introspect and confirm any of this. Both arguments run by elimination, which is unusual in this material and worth noticing.
Then comes the move that makes the scheme worth a page, and the reason for a third part isn't awareness of awareness. It's memory.
If this Bhāga which is aware of the darśanabhāga were lacking, there would be no memory or recollection of the mind and its activities, just as there is no memory or recollection of images that have never been perceived.2
Follow it. You can recall having seen the blue. Recall needs the seeing itself to have been registered — a bare seeing that nothing took note of would leave nothing behind to remember, in exactly the way an image you never perceived leaves nothing. So something must have recorded that perceiving occurred, and that something is a third part.
That's a functional argument for a component of mind, derived from a capacity nobody disputes. It appeals to no special experience at all, which puts it well ahead of most of what surrounds it.
The three then get stated cleanly: the image, nimittabhāga; the perception of the image, darśanabhāga; and "the knowledge or awareness that perception of the image has taken place," svasaṃvittibhāga.2 In the act-and-fruit vocabulary the tradition prefers, the image is that which is perceived, the second is the act of perception, and the third is "the fruit of the act of perception."2
The third also gets a structural job. It's the dravya, the "thing" — described as "the infrastructure and the essential nature of the first two Bhāgas, and which is consciousness itself."2 The other two are aspects; this is what they're aspects of.
There's a related distinction in the same passage that keeps the parts from blurring. Consciousness and its associates share a supporting basis, and the images they perceive closely resemble each other without being the same. Their actions differ: the action of consciousness is to perceive, and the action of sensation is to feel. They're equal in number — one consciousness, one sensation — and different in character.2
Now the obvious objection, and the text raises it against itself before anyone else can:
If a fourth Bhāga is lacking, by what will the svasaṃvitti, in its turn, be perceived and corroborated? The svasaṃvitti, being a Bhāga of the mind, must be corroborated, as the darśana is corroborated by the svasaṃvitti.3
That's the regress in one sentence. If knowing has to be known, then knowing-that-you-knew has to be known too, and there's no visible place to stop.
A second argument runs in the act-and-fruit frame. The svasaṃvitti, being an act of perception, must have a fruit. Could the darśana be that fruit — could the loop just close backwards? No, and the reason is technical rather than convenient. The darśana "is not always characterized by 'direct or sense inference'"; it sometimes works by correct inference and sometimes by erroneous inference, so it can't serve as a corroborator of anything.3
Hence a fourth part: the svasaṃvitti-saṃvitti-bhāga, "the Bhāga which corroborates the self-corroboratory Bhāga," and it's specified as belonging to the direct-inference category.3
Here's the part that makes this engineering rather than an arbitrary halt.
The third Bhāga… can take as its objects both the second, darśanabhāga, and the fourth, svasaṃvittisaṃvittibhāga. The fourth Bhāga… takes only the third as its object; it does not take the second which is already taken as its object by the third. Both the third and fourth Bhāgas belong to the 'direct inference' category.3
Three takes four as an object. Four takes three as an object. The two hold each other up, so neither needs anything above it, and the regress ends.
That isn't a stipulation that the buck stops here. It's a loop closed by mutual objecthood, and it works because both members are restricted to direct inference — so neither can be the source of an error the other would have to catch.
Notice what the scheme doesn't do: it doesn't add a fifth to check the fourth. And the reason it doesn't need one is the same reason the second couldn't close the loop, which is reliability. The regress is stopped by a pair of reliable parts checking each other rather than by a foundation underneath them.
There's a tidy structural summary attached. Of the four, the first two are external and the last two internal. And the first is only an object of perception; the other three are both subject and object.3
Having built to four, the text immediately shows that the count is a convenience.
The four Bhāgas may be comprised in three, because the fourth may be included in the third. They may be comprised in two, because the last three… can be comprised in the 'perception' aspect. Alternatively, they may be comprised in one, because their substance is not different.3
Four, three, two, one — the same object analysed at four grains. The Pramāṇasamuccaya stanza says the same thing about the first three: "the substance of these three is not differentiated."2 And the Laṅkāvatāra supplies the terminus: "Being attached to itself, the mind manifests into what seems to be an external sphere of objects. What it perceives has no real existence. Hence it is affirmed that there is nothing but Mind."3
Which reframes everything above it. The four parts aren't four things. They're four descriptions of one substance, produced by asking four different questions of it — and the scheme says so itself rather than leaving a reader to work it out.
The scheme makes a prediction about what happens when somebody tries to find these parts by looking, and it's worth running, because the prediction is unusual: it predicts you'll fail at a specific point, for a specific reason. Sit down with a cup. The first part is easy — the cup as it appears, which isn't the cup but the appearance of it. The second is harder, because attention slides straight back onto the cup every time you reach for the looking itself; most people get a flicker and lose it.
The third is available, and there's an honest test for it that doesn't depend on catching anything in the act. In ten minutes you'll be able to say I was looking at a cup. You couldn't say that if nothing had registered the looking. So the third part is demonstrated by the memory rather than by the introspection, which is the same argument the text used, run on yourself.
Then try for the fourth: what knows that? And here you hit a wall. Every attempt returns another instance of the third rather than a distinct new thing. You can do it ten times and the tenth feels exactly like the first.
The temptation is to read that as failure — as evidence that you're not subtle enough, or that the scheme is describing something above your level. It isn't either. It's the argument working. Three and four take each other as objects and have no separate content, so there's nothing it would feel like to find the fourth as a fifth kind of thing. Hitting the wall is what their substance is not different means when you do it instead of reading it.
Which is the case's real yield, and it generalises past this scheme. A model that predicts where introspection will stop producing new objects, and turns out to be right about the location, has told you something. A model whose parts you can always find one more of has told you nothing, because you were always going to find one more.
Forget the four for a moment and take the third one home, because it's the part you actually use. Every time you catch yourself having been somewhere else for ten minutes — the walk you don't remember, the page you read four times — that's the registration reporting late. It was off, and its coming back on is the thing you noticed. Nobody experiences an absence while it's happening; you only ever experience the return.
So use recallability as the measure of a sitting instead of calm. At the end of twenty minutes, before you get up, say what happened — not how it felt, what happened. There were about four minutes of settling, then a long stretch on the breath, then the argument from Tuesday arrived and stayed for a while, then something quieter. If you can produce that, the third part was working. If the whole twenty minutes is a smooth blank with a verdict attached — that was a good one, that was restless — you have a verdict and no record, which is a different thing and a much less useful one.
That reframes what to expect early on. Whatever else meditation trains, it trains the registration, which is why the first reliable result isn't peacefulness. It's the ability to say what happened. People give up at six weeks because they were watching for calm and calm is slow; the thing that actually improved in those six weeks was the account they could give afterwards, and nobody told them to look at it.
And carry the collapse with you as a brake. When the analysis starts producing exciting new interior objects — a watcher behind the watcher, a deeper awareness underneath that — remember that the scheme's own answer is that their substance isn't different. Finding more parts isn't progress. Finding where the parts stop being different is.
[POPULAR SOURCE]. Range read: pp 234–236. The primary text arrives as Bodri citing Wei Tat's translation of Hsüan Tsang — a properly located citation of a named translation, and one of the better ones in the book. Full chain in the footnotes.
The arguments are the strongest thing here, because they run by elimination and none of them asks the reader to introspect and agree.12 The regress is raised by the text against itself and stopped by a stated mechanism, and the four-to-one collapse comes from the same passage, so the scheme never pretends its parts are separate things.3
The unclosed tension is about reliability. The darśana is excluded as a corroborator because it sometimes works by illogical inference — an argued exclusion. The third and fourth are simply asserted to belong to direct inference.3 It is argued for the part kept out and stipulated for the parts kept in.
A second, on transmission: the passage is printed twice, in two translations, with the reader asked outright whether they can understand it.1 That is evidence for the same page's claim that the terminology has stopped carrying its meaning.
Scope limit: the scheme describes how perception authenticates itself, not an attainment. It's attached to the path anyway, and the stages authentication gets applied to are the ones the source's own footnote 518 calls conventional.4
Open question. If three and four are distinguished only by which takes which as an object, is the fourth doing work the third couldn't do alone — or is it a formal device for closing a loop that would otherwise look open?
Two instruments from one tradition disagree about what verification even is, and this material uses both without noticing. The ten-question checklist from the Source Mirror assesses a practitioner without once asking what they experienced — every question is about continuity, imperturbability, or meeting the teachings without fear. This scheme is entirely about the structure of experiencing, and its whole content is an anatomy of the inner act.
Follow the first and inner evidence is worthless: what you felt isn't the question, and a person who reports remarkable states has told you nothing. Follow the second and the inner act is the only thing there is to analyse, and its self-corroboration is a considered position defended on principle.
Nothing reconciles them, and the disagreement isn't small. One says verification must come from outside the experience; the other says, for this class of claim, there's no outside. Both are in the same corpus, applied to the same practitioners, and the corpus never puts them in the same room.
History — The Multi-Agent Corroboration Method. Two letters reach a commander from two men who don't know the other exists, sent at different hours by different couriers, worded differently, pointing at the same thing. That overlap is evidence precisely because it couldn't have been arranged. And the manual names the failure mode exactly: the moment agents become aware of each other they stop reporting and start coordinating, and hand you one clean story — a clean story from people who have talked being worth less than two messy stories from people who haven't. The third and fourth bhāgas are the coordinated case at its theoretical limit. They corroborate each other, they take each other as objects, and the text states plainly that their substance isn't different.3 By the intelligence standard, that corroboration is worth nothing at all. And yet the scheme isn't making a mistake, which is what the pairing exposes: the two cases are asking corroboration for different products, and only one of the products admits independence. Grain moving west is outside both agents, so independent access is possible in principle and agreement carries information. Whether this mind knew something has no outside — nothing that isn't the mind has access to it — so every conceivable corroborator is a component of the system under examination. Mutual grounding isn't a design flaw here; it's the only design available. Which yields a rule that sorts a large class of disputes and is uncomfortable in both directions. Self-knowledge is categorically weaker than knowledge of the world, and no amount of internal machinery upgrades it: build two parts or forty, all of them are inside. So how elaborate a verification scheme is measures the difficulty of the problem rather than the reliability of the answer, and a very sophisticated account of how introspection checks itself should lower your confidence rather than raise it. The practical test: ask what your corroborators would need to be independent of, and whether they can be. Where they can, insist on it. Where they can't — introspection, self-report, claimed inner change, attainment verified by the attained — no procedure buys what independence buys, and the honest move is to stop treating internal verification as a weaker version of the external kind. It's a different kind, and these four parts are the most careful statement of it anybody has produced.
Sharpest implication. Corroboration is worth exactly what the independence of its sources is worth, and for any claim about your own knowing, independence is impossible in principle — every possible corroborator is a component of the thing being corroborated. So a very elaborate account of how self-knowledge checks itself is measuring the difficulty of the problem, never the reliability of the answer, and should move your confidence down rather than up.
Generative questions.