A retried write checks the owner before calling itself a no-op

An update to a node held by another part of the mesh can be answered with "that never applied" while the shutting-down owner has already told every reader about it. The retry then saw its own value already present, decided there was nothing to write, and reported success — for a change that reached no store.

That was closed for one form of "nothing to write". A second form remained: when the value was rebuilt rather than reused, it compared as different and only turned out to be identical once written out. Both forms now ask the owner what it actually holds. If the owner has the value the success is real; if it does not, the retry writes it.

The phantom-base investigation records the mechanism, why a mirror can never answer the question on its own, and what a test must observe for a lost write to be reported as one.

Reconnecting…
The server was updated. Reloading the page to pick up the latest version.