REPORT + STATUS: instruction arc close-out

Operator line for the one change with real-world consequence: the runbook that says which machine
may be wrecked claimed the HP box keeps no off-site copy, and it keeps two. Checked against the
machine, not corrected from memory.

STATUS.md

@@ -76,6 +76,15 @@ failed until someone walks it end to end with no help from us.
 
 ## What shipped recently
 
+- **2026-08-06 (later)** — **The page that says which machine may be wrecked had one wrong sentence,
+  and it has been checked against the machine rather than corrected from memory.** It claimed the
+  spare HP box keeps no off-site copy, and used that as the reason to aim risky backup tests at the
+  other one. The HP box **does** keep an off-site copy — two of them, the newest from 4 August, in
+  its own space on the off-site server. The sentence was true when it was written and went out of
+  date on 23 July. It matters because a belief that a machine holds nothing is how a destructive test
+  lands on one that does. Also today: the notes the assistant keeps had three statements that were
+  simply untrue (including that the demo boxes were still away) — those three are fixed, the rest are
+  now flagged automatically rather than rewritten by hand. *(R-229, R-230)*
 - **2026-08-06** — **The assistant's own working notes were on one machine with no copy anywhere.**
   Everything the assistant has learned about this system over months — where things live, which traps
   cost us an incident — sat in a folder on the build machine that no backup touched and no repository