Seven women go to the fire. Six of them are deluded by Śiva's māyā, approach it, and come away pregnant. Their husbands throw them out. The text's line on it is "Those six wives saw that they had transgressed."
Seven went. Six were deluded.
Arundhatī was told in advance. The text says so, in passing, in a subordinate clause — instructed by Girīśa — and she is the one who tries to restrain the other six, and she is the one whose life does not come apart.
Same fire. Same night. Same god. One woman briefed, six not, and the outcome tracks the briefing exactly.
That is a control group, sitting in a devotional text from the eleventh century, and nobody sets it up as one. Not the narrator, who has just written the cleanest natural experiment in either volume and moves straight on to the next thing. Not the translator, whose notes are attentive on both sides of the passage. Not Volume 1's build, which had the six wives on a page and did not have the seventh.
First wire, and it's the obvious one. Disclosure is the whole variable. Not virtue — the text never says Arundhatī was more devout. Not resistance — she doesn't resist, she knows. The six aren't punished for wanting something; they're punished for having been operated on without being told, and the text prices that as their transgression.
Second wire, and it's the one I can't let go of. The narrator didn't need Arundhatī. The story works without her. Six wives, māyā, disgrace, done. She is a surplus detail — and surplus details in a formulaic text are the ones that came from somewhere.
Which raises a question the build couldn't close: is she a fossil? A remnant of a version where the exemption was the point, sanded down until only the clause survived? Or is she doing something I'm not seeing, and the "control" reading is my instrument talking?
Third wire, and this is where it stops being about a Purāṇa. I found her because the liability rule I was building needed a case where the impairment was disclosed. I went looking for a counter-instance and she was already there, one clause deep, in material this vault had read.
⚑ The thing I want to sit with: how many controls are lying in read material, invisible because nobody had a hypothesis that needed them?
Six confirmations were sitting in this build's etiology notes for a week — nine out of nine, injury-derived, collected because they were injury-derived. The sweep that broke it took four minutes. Arundhatī is the same shape running the other way: a disconfirming case, already read, already filed, waiting for a claim strong enough to make her matter.
Not go back and re-read everything. That's the useless version of this. The useful version is smaller and meaner: when you state a rule, immediately ask what its exempt case would look like — and then check whether you've already met one.
liability-and-the-undisclosed-impairment, where she already is, but read the other way round — as method rather than evidence.Filed from the Shiva Purana Volume 2 DEEP INGEST, 2026-08-11. [PERSONAL] on the third wire — the generalisation about controls in read material is mine, not the text's.