The posts, revised against the code rather than against memory

The posts, revised against the code rather than against memory Every claim now points at something checkable, and several got narrower in the process. Verified before committing: `verify_failures` is real and sits in `0002_storage.sql` — the earlier draft said the scrub columns had no writers at all, which overstated it; `Capabilities` matches the post field for field, including the note that SeaweedFS echoes a storage-class header without honouring it; `Key::original` builds the digest-prefixed path the post prints; and `const PAGE = 60` is where the asset page's window claim comes from. Two numbers are now hedged where the first draft asserted them. The scrub counts — 608 missing at one checkpoint, 609 at another — are observations from different checks during a live incident rather than a census, and the post says so. The browser flakiness figures are labelled as one historical diagnostic against a 410-test suite rather than a standing property of the current one. Post 6 also admits the asset page does not fetch windows past its first 60 assets, which limits the claim its own title makes. Leaving that in is the point: the series argues that a system's silence about its faults is not evidence there are none, and a post that exempted itself would be making the argument it warns against.