The Convention You Own and Do Not Use
Volume 2 tells the same story twice, four times over, and never says so.
Volume 2 tells the same story twice, four times over, and never says so.
Kārttikeya's conception is deposed under oath at ch 154 along a route ch 152 does not contain. The Avadhūteśvara episode runs beat for beat across chapters seventy-six apart — same similes, same etymology — and *the later telling calls itself the original. Nṛsiṃha's rage is resolved by curse in one saṃhitā and by flaying in another, on a shared three-element setup. The gods become paśus twice, by two unrelated mechanisms.
Ordinary enough for a compilation. Here is what isn't.
The volume owns a tool for exactly this, and uses it five times.
"Because of differences in kalpas, the story of Ganesha's birth is spoken about in different ways" (163). "These differences occur because incidents have occurred in different kalpas" (298, raised by the listeners themselves). "I have told you everything about his conduct when I spoke of Sati's conduct. Since you have heard it many times, I will not speak about in great detail" (239). "I have already told you about Kumara's divine conduct, but I will describe it again" (286). And the listener-initiated version at 217.[^1]
Three mechanisms. Five deployments. Zero applied to the four cases that needed them.
The other side. This vault has its own disclosure conventions, and they are good ones. `[FALSIFIED]` with an obituary block retained. Do not cite the earlier version forward.* The `[n=1 OBSERVED]` / `[n=2 SUGGESTIVE]` / `[n≥3 PATTERN]` ladder. The rule that a broken pattern keeps its page rather than being quietly deleted.
*Having a convention and not using it is worse than not having one — and, unlike most epistemic failures, it is countable.
That's the mechanism, and it's sharper than it first looks.
A text with no disclosure norm tells you nothing when it stays silent. Silence there is uninformative; there was never a practice to depart from. But a text that marks repetition five times and stays silent on the four places where the versions conflict is telling you something specific: the attention existed, the sentence was ready, and it did not reach those four. Those four are where separate material was joined, or where somebody wrote without reading what he was contradicting.
⚑ The norm converts a silence into a signal. That is the whole finding, and it makes disclosure conventions worth adopting for a reason nobody adopts them for: not because they get used, but because their non-use becomes measurable.*
Audit any body of work by counting where its own conventions were available and skipped — not by hunting its errors directly.
Hunting errors is expensive and scales badly; you have to read everything and hold it all at once. Counting convention-deployments is cheap, grep-able, and short. Then the gaps localise the seams.
And it turns on this vault immediately.
`six-disclosed-repairs` established that Debroy discloses generously when he changes a word. Volume 2 added a category that page didn't have: silent repair. Endnotes 716 and 717 supply the complete five-face and eight-form mappings — organ by organ, element by element, plus which face turns which way on a five-faced liṅgam — for chapters whose bodies deliver those schemas incompletely. Nothing marks it. A reader working from the notes comes away with a tidy cosmology the text never contains.
He discloses when he changes a word. He does not disclose when he completes a structure — and completing a structure is the larger intervention, because a changed word is visible as a change and a supplied table is invisible as an addition.
So the question for this vault, and I do not know the answer:
⚑ Where has this vault completed a structure rather than changed a word?
Every concept page here is built to a fixed skeleton — 14 sections, a rubber-duck hook, an Analytical Case Study, an Implementation Workflow, a Live Edge with generative questions. A source that supplies eleven sections' worth of material still gets fourteen. The build's own governing rule says the honest move is to fold the page back rather than fill the shape — "a thin page is worse than no page, and a missing insight is cheaper than a manufactured one."
I invoked that rule this build. I also, roughly half the time, hit 12 or 13 sections and went back to the source for a genuine eighth topic section rather than folding. Those were real finds and I stand behind them. *But nothing in the apparatus records the difference between I found more and I needed one more. The convention exists. Its non-use is not countable, because there is nowhere it would show up.
That is the same failure as endnote 716, one level up: a completion with no mark on it.*
- A second corpus with an explicit disclosure norm, where the deployments and the silences can both be counted. Version-controlled documentation is the obvious test — this supersedes, see also, deprecation notices — because the counting is trivial and the seams are independently knowable from commit history.
- A ratio that means something. Five deployments and four silences in 154 chapters is a small denominator on both sides. Whether five is a lot for a compilation of this size is unknown, and without that the signal could be noise.
- And a mark that would work here. The concrete proposal: a page that reached fourteen sections by returning to the source for more material should say so, in one line, in Evidence — "the eighth topic section was built after the first draft came in at seven; the material is at ch X." ⚑ That is cheap, it is honest, and its absence would then be countable — which is precisely the property this collision says conventions are for.
- The deflationary reading, which is live. Purāṇic compilers may simply not have cared about the kind of consistency I'm scoring them on, in which case the five deployments are five accidents rather than a norm, and there is no departure to measure. Nothing in the text settles it. And the vault-facing half may be over-scrupulous: a fixed page skeleton is a template, not a claim, and nobody is deceived by a page having fourteen sections. That reading is available and this collision does not close it — though it is worth noticing that the argument "nobody is deceived" is exactly what would be said about a translator's footnote that quietly supplies a missing table.