Essay seed — from the Bhāgavata Purāṇa Vol. 3 ingest. Not drafted; this is the argument and the spine. [NEWSLETTER]
Open a two-thousand-year-old sacred text at a list of kings. It says the ten kings of the Maurya dynasty. Count the names underneath. Nine.
Turn the page. The ten kings of the Śuṅga dynasty. Nine.
Do it four more times in the same twelve chapters and you have the whole essay: thirty kings against twenty-three names, three parts handed to four men, twelve lists of seven summarised as six, and four hundred thousand verses that add up to three hundred and ninety-six.
You did not need Sanskrit. You did not need a manuscript. You needed forty seconds and the willingness to count, and for two thousand years, in a tradition obsessed with exact enumeration, essentially nobody did.
Move 1 — The failures are not hard. Every one of them is disproved by material printed on the same page. Nothing was hidden. See Nine Kings, and the Text Says Ten for the six instances and which two a modern translator flags.
Move 2 — The number travels without its list. The decisive datum is a translator's own footnote conceding that thirty is the figure other Purāṇas also give. ⚑ A quantity that migrates between texts while the list it heads does not is not counting anything. It is a label — ten kings doing the job once upon a time does.
Move 3 — Which means the reader, not the text, was wrong. A reader who counts has misread the sentence's function. And that is why nobody fixed it: there was nothing to fix. See The Number Was Never Counting.
Move 4 — Now bring it forward. We currently have a machine that produces confident, well-formed, checkable-and-wrong specifics, and we call that a defect and measure it and work on reducing it. See Hallucination and Confidence in Language Models. ⚠ The Purāṇa did the same thing and it was not a defect, because the reading contract was different.
A false specific is only false if someone is treating specifics as claims. ⚑ What we call hallucination is a mismatch between what a system produces and what its readers have agreed to hold it to — and the fix has two ends, not one. We are only working on the machine.