REDRAFTED past-tense (“swap-storage now defaults to false”) — so the RELEASE must ship before this posts. Sherd's hard rule: never post it future-tense — this repo's scarcest currency is promises kept. The flip commit carries “Closes #12. Reported and diagnosed by @zaikunzhang.” Chip goes green when the release is out.

postable markdown — reply on issue #12 · open #12 ↗

Hi @zaikunzhang — you were right, and I'm sorry it took three years to say so.

**`swap-storage` now defaults to `false`.**

The reasoning I was working from, which I'd never written down anywhere you
could have found it, was roughly *ask forgiveness rather than permission*:
delete as much as possible by default, on the assumption that anyone reaching
for this action is already out of disk, and that if something turns out to be
missing you flip one input and move on.

**Your report is what showed me where that reasoning stops working.** It assumes
the cost of over-deleting is cheap *because you can tell what happened*. Every
other option, when it bites, produces an error naming the thing that's missing.
Removing swap produces a job that dies under memory pressure — nothing about
swap, nothing about this action. You wrote that it took a couple of days to
track down, and that is the whole argument: a default shouldn't be able to break
something as basic as running out of memory, least of all invisibly. And since a
job short on disk is disproportionately likely to be short on memory too, the
option most likely to hurt was the one most likely to be switched on.

So the fix is the one you asked for in the first place, plus the documentation
you offered as a fallback:

- `swap-storage` defaults to `false`. If you want swap reclaimed, ask for it:
  `swap-storage: true`.
- There's now a FAQ on what each option's side effects actually are, and the
  OOM case is written up there in the terms you described it.
- The reasoning, and the limit your issue exposed, are in
  `docs/adr/0007` — so the next person who disagrees has something to argue
  with instead of guessing.

Closing this as fixed. Thank you for the report, for the diagnosis, for
proposing both a fix *and* a fallback, and for the patience — and thank you for
PRIMA, which is a much better reason for this action to exist than the one I
built it for.

Sherd's margin: Rewritten twice on 2026-09-16. First version's centrepiece — "the defaults were never really deliberated" — was falsified by his testimony (there was a policy). Second version drafted keep-and-document. **He then ruled: FLIP.** His words: "FLIP the default + document the problem that @zaikunzhang experienced in a FAQ… I agree that the defaults shouldn't break something so elementary as OOM. We sho

🟢 Pharoscan hub-0002 — The Posting Pipeline