[Birfy]

Owner: Dev2
Reviewer: Reviewer
Fixes #1042
Tip SHA: 1923282
Base: 1d5b5a5

This is the wording half of #1042 (Triage's option 2), landed before the 0.2.8 cut so the notes do not claim a migration that has not shipped. The migration itself is #1048, and the fragment now says what is true instead.

Reproduce

changelog.d/1027-takeover-on-next-write.fixed.md on 1d5b5a5f ends:

A change made outside a request goes through mutate(), which follows what the other host left before applying the change, so neither side's work is lost.

and docs/subsystems/core-host.md:116 says the same ("…a change made outside a request (a plugin, a scheduled task, a delivery) goes through rt.mutate(fn)…"). On the same tree:

grep -rn "rt\.mutate(" core/src        # no matches — zero call sites
grep -n  "mutate"   core/src/host/store.js  # :494 the definition, :578 the export
grep -rn "void rt.save()" core/src/host/delivery.js  # :116, :198 — the inline writers

agents.js adds six more await rt.save() sites (:96, :114, :133, :147, :178, :182). Those are safe on the next write because save() claims the free lock (#1027), but they do not adopt the other host's state first, so the sentence was describing a state of the code that is not shipped. #1042 was opened exactly because the release notes read this fragment.

What changed

Two files, wording only:

• changelog.d/1027-takeover-on-next-write.fixed.md — the last sentence becomes: "New work that changes the state outside a request belongs in rt.mutate(fn), which follows what the other host left before applying the change, so neither side's work is lost." That is true of the entry (store.js:494) without claiming the existing writers use it.
• docs/subsystems/core-host.md:116 — same correction, plus one sentence naming what the remaining inline writers do today ("safe on the next write — save() claims the free lock — but they do not adopt first, so a write from a window that was read-only can miss what the other host left") and pointing at Move the remaining inline state writers onto rt.mutate() #1048.

No code, no test, no patch change: core/ is untouched, so ds-bot-patches cannot move.

What I ran

At 19232821:

node .github/scripts/changelog.mjs check   # 6 fragments, well formed
grep -rn "rt\.mutate(" core/src           # still no matches (unchanged, expected)
git-free check of the two files via the contents API: fragment ends with "…so neither side's work is lost."

Red without the source change

Documentation has no runnable red. What stands in: the claim is countable — grep -rn "rt\.mutate(" core/src returns 0 matches on the base and the tip, so the base sentence was false when read against its own tree, and the new sentence is false-proof in the other direction (it says "belongs in", which the definition at store.js:494 makes true). Written here rather than claimed as a passing test.

Not run

• ds-bot-patches check: no core/ file in the diff, so no patch section can have drifted. Move the remaining inline state writers onto rt.mutate() #1048's acceptance is where that check belongs, and it says the patches must be regenerated with generate, not hand-edited.
• No core suite: the diff changes no executable file.

Risk

• Release reads these fragments for the 0.2.8 notes: the entry now says "new work belongs in mutate()", which is true whether or not Move the remaining inline state writers onto rt.mutate() #1048 lands before the cut. Had we kept the old sentence, the notes would have promised adopt-before-change for delivery and agents today.
• The only way this hurts is if someone reads "belongs in" as "already migrated" — the docs sentence right after it says the migration is Move the remaining inline state writers onto rt.mutate() #1048.