Read Volume 1 end to end and the same question gets answered twice, differently, seven times. Every occurrence is annotated. No note anywhere says the other one exists.
| the disagreement | where | |
|---|---|---|
| 1 | Body sheaths: seven or eight — the extra term is Prakṛti | 2(1) / 2(10), nn. 442 and 673 |
| 2 | Kinds of primal ignorance: four or five — and different names | 3(12) / 3(20), n. 914 |
| 3 | The cosmic body mapped three incompatible ways, and 2(5) admits a fourth exists | 2(1) / 2(5) / 3(13) |
| 4 | Two avatāra catalogues, differently constituted and ordered | 1(3) / 2(7) |
| 5 | Two causal accounts of the Jaya–Vijaya curse, minutes apart, both spoken by Viṣṇu | 3(16) |
| 6 | Two etymologies for Parīkṣit, both kept, both used | 1(12), n. 278 |
| 7 | Two tattva schemes — 4+16+5 against 24+1 with a four-fold internal organ | 2(9) n. 650 / 3(26) |
[n≥3 PATTERN] — seven instances is well past the threshold, and the vault's constraint on register applies: this is a count, not a theory of the text.1
They are not all the same species, and lumping them flattens the finding.
Three are counts. Seven sheaths or eight; four ignorances or five; two tattva schemes. A number is given, and elsewhere a different number is given for the same object.
Two are inventories. The avatāra catalogues and the cosmic-body maps — same category, different membership, different order.
One is causal. Viṣṇu gives two incompatible reasons for the same curse, in one conversation.2
And one is etymological. Parīkṣit's name derived two ways, both retained, both used in the narrative.
The count-disagreements are the easiest to dismiss as transmission noise. The causal one is not, because it is two explanations from one speaker in one scene, and the volume needs neither to be true for the story to work.
Six of the seven are what you would expect from a compiled text: a number drifts, a list is reordered, an etymology accumulates.
The seventh does something else, and it is the sharpest item in the census.
Both tattva schemes total twenty-five. Endnote 650 at 2(9) gives 4 + 16 + 5. Chapter 3(26) gives 24 + 1, with a four-fold internal organ whose functions are specified — doubt, conclusion, pride, remembrance.3
The totals agree and the membership barely overlaps.
Which is the reverse of a transmission error. A number that drifts is evidence of copying; a number that holds while its contents change is evidence of a convention being honoured.
And it produces a general observation worth keeping: where a total is fixed by convention it carries no information, and the membership carries all of it. Twenty-five is what a Sāṅkhya scheme is called. What it contains is the actual claim, and a reader who checks the total will find agreement where there is none.
That inversion is the census's most useful item, and it is the one nobody would find by counting.
Every one of the seven is annotated. Debroy glosses the seven-sheath passage and the eight-sheath passage. He glosses four ignorances and five. He supplies both Parīkṣit etymologies. He gives the 4+16+5 scheme at 2(9) and annotates the 24+1 scheme at 3(26).
He never says that the other one is there.
Each note fits its own local passage and is correct for it. The apparatus is locally accurate and globally silent — which is not carelessness, because catching these requires holding a whole volume in mind while writing a footnote, and that is a different job from translating.
But the vault has this behaviour on record from another book by the same translator. The Śiva Purāṇa build logged the five kinds of jñāna and the ten glossed 273 notes apart with no recognition.4
So the pattern holds inside the Bhāgavata at [n≥3], and across the Debroy corpus the translator behaviour is now [n≥3] as well.
Three restraints, stated because the finding is easy to over-drive.
It is not evidence of multiple authorship. It is compatible with that and with several other explanations, and this build has read one translation of four Skandhas. Nothing here settles composition history. [SPECULATIVE — no claim about redaction is made.]
It is not a charge against the translator. A local gloss that fits its passage is doing its job. The observation is about what an apparatus organised per-passage cannot see, which is a structural property of footnotes rather than a failure of attention.
And it is not a claim that the text is confused. Six of the seven are the ordinary condition of any long compiled work. The finding is the seventh — the tattva case, where the total is preserved and the contents are not, which is a positive act rather than a drift.
The practical consequence is worth stating plainly, because it is why the census matters at all.
Nobody reads a Purāṇa the way this build read it. A reader arrives at 2(1), learns there are seven sheaths, and has no reason to return. A reader who arrives at 2(10) learns there are eight. Both readers are correctly informed and neither is in a position to notice.
And the apparatus reinforces it. Each note tells you what this passage means. The one thing a footnote cannot tell you is that it disagrees with a footnote four hundred pages away.
Which makes the disagreements invisible by construction — not hidden, not suppressed, but structurally unavailable to the only reading mode the object supports.
The ignorances are worth following in detail, because they show what the census looks like at close range.
At 3(12), Brahmā's first creation is a set of four kinds of ignorance, and he is ashamed of it. At 3(20), the same creation is a set of five.6
The counts differ. That much is an ordinary drift.
The names differ too, which is not. A four-item list and a five-item list that shared four names would be a list with an addition. Two lists with different membership are two schemes, and the volume presents both as what Brahmā created first.
And the note at 3(20) glosses the five without mentioning the four.6
Now put that beside the sheaths. At 2(1) the count is seven; at 2(10) it is eight; the extra term is Prakṛti, and both are annotated.7 There the delta is a single identifiable item, which makes the disagreement recoverable — a reader who holds both can see exactly what was added and reason about whether it belongs.
Two disagreements, two different repair costs. The sheath case is one term and can be reconciled by anyone. The ignorance case cannot, because there is no mapping between the lists to argue about.
Which is why the census sorts by species rather than counting instances. Seven is the headline and it is the least useful number in the finding. What matters is that the seven divide into drifts you can repair, inventories you cannot, one causal contradiction inside a single conversation, and one preserved-total substitution — and only the last two are evidence of anything beyond ordinary compilation.
You have a body of material that has grown over time — documentation, policy, a codebase's comments, an institution's guidance — and you suspect it contradicts itself.
Sweep by category, not by chapter. These seven were found by asking what does this text say a number is across the whole volume, rather than by reading in order. Contradictions are invisible to sequential reading and obvious to a categorical pass, and almost nobody runs the second kind.
Check the totals against the contents. The tattva case is the one that would survive any audit that compares figures, because the figures agree. Where a headline number is conventional, verify membership; the agreement of the total is the thing that will stop you looking.
And distinguish drift from convention. A count that varies is probably noise. A count that holds while its definition changes is a decision somebody made, and it is the one worth chasing — because someone knew the number mattered and preserved it while changing what it referred to.
The strongest evidence that this is a property of the text rather than of the reading is the causal case. Viṣṇu gives two incompatible reasons for the same curse, minutes apart, in one conversation, and neither is withdrawn.2
Tension — the apparatus is locally accurate at every one of the seven and globally silent at all seven.5 Whether that is a limitation of footnotes or a choice is not determinable from the volume. Preserved.
Tension — six items are consistent with ordinary compilation and the seventh is not. The census treats them as one pattern and they may be two. Preserved unresolved.
Open question. The volume states at 2(5) that a fourth cosmic-body map exists. Is that map anywhere in the four Skandhas, or is the admission all there is?
The developed Vaiṣṇava commentarial tradition has instruments for exactly this — tātparya, harmonisation, the assumption that apparent contradiction indicates a reader's limitation rather than a text's — and it applies them with skill.
This build does not, and the reason is CLAUDE.md's first hard constraint: never silently resolve contradictions.
What is worth recording is that harmonisation and census are answering different questions. A commentator asks how can both be true. This page asks how many times does this happen, and does the pattern have a shape.
The answer to the second turns out to be more interesting than the first would have been, because the tattva inversion — total preserved, membership replaced — is a finding that harmonisation would dissolve rather than notice. Reconciling twenty-five with twenty-five is trivial. Noticing that the reconciliation is empty requires counting.
With Cognitive Priming and Memory Coloring. That page holds the finding that immediately preceding context shapes how ambiguous material is encoded — that the same input is understood differently depending on what was in mind a moment before, and that the colouring is invisible to the person it happens to.
The apparatus is that mechanism operating at the level of a scholarly practice. Each note is written with one passage in mind. The passage primes the reading; the reading produces the gloss; the gloss is correct for the passage and is silently shaped by it.
What the pairing produces: priming research is almost always about a subject who is unaware of the influence. This case adds a structural feature the laboratory version cannot have — the priming context is the object of study itself, and the practitioner's job requires them to attend to it closely.
Which means the better a translator is at holding one passage in mind, the more locally accurate and the more globally silent the apparatus becomes. Expertise increases the effect rather than reducing it.
And it identifies the only available countermeasure, which is not vigilance. Nobody detects their own priming by trying harder. What catches these is a second pass organised on a different axis — by category rather than by sequence, so that the seven-sheath passage and the eight-sheath passage are in mind at the same time. That is a different piece of work from annotating, and no amount of the first produces it.
Sharpest implication. Six of the seven disagreements are what any long compiled text produces. The seventh is not: both tattva schemes total twenty-five and share almost no membership, which means the number was preserved while its contents were replaced — the signature of a convention being honoured rather than a figure drifting. Where a total is fixed by convention it carries no information and the membership carries all of it, and a reader who checks the total will find agreement where there is none.
Generative questions.