docs(drift): a remedy tested in the direction that works, parked

[JulienMartel]

What

One entry appended to docs/drift.md's Seen once, not yet a row. Numbering untouched — still thirty-three.

Why

Re-measuring the register's cross-room S2 against haus capture / haus revert-settings turned up a shape worth keeping: a remedy that works in the direction anybody would test it and fails in the direction anybody would need it.

The restore is defaults import, which merges, so it returns a key the snapshot held and cannot remove one added since. Measured on a guest running haus ef6e808d: com.apple.dock tilesize captured at 48 and changed to 67 comes back to 48; the same key absent at capture and written to 67 is still 67 after ✓ com.apple.dock restored prints.

Anyone verifying the verb reaches for a setting they had already chosen — the working half. The half that fails is the settings a rebuild adds, which is the only reason the verb exists.

Verify

The entry disposes of row 27 by name: row 27's remedy produces nothing at all and is falsified inside the same rev by the same repo, where this one half works, passes the obvious test, and nothing in the repo had ever contradicted it. The haus-side corrections are hausfold/haus#765 and the site's are hausfold/hausfold.co#378.

Watch out

• Rebased onto main after docs(drift): a mechanism inferred on the wrong axis, parked #617 squash-merged; the earlier entry in that section is now main's, not this branch's.
• Its Caught by asks a question about the empty cell in a before-state, which is the transferable part: a fixture written from the happy path never has one.