FABRIC-3.md: close D.7's persistence prerequisite

Hermes and Artemis are now session-less fleet foundation, verified live on all three architectures (commit 52eb1bb) -- the real, load-bearing prerequisite D.7 itself identified is done. The rest of D.7 (message-only birth flow, WIREBIND/MSGMIGRATE reconciliation) remains open, not touched here.

FABRIC-3.md

@@ -2527,7 +2527,16 @@ persistently alive today. Every boot this session confirms she's birthed for her
 migration self-test, then explicitly killed (`KILL: Artemis dead`) before the REPL ever
 reaches `ok>`. For any message to reach a running Artemis instance, she has to actually stay
 born — a real, new requirement this direction depends on, not something `HOTPLUG`/`WIREBIND`
-ever needed before.
+ever needed before. **CLOSED 2026-08-28**: `kernel_main.c`'s item 4.2/4.6 self-test blocks
+(birth → exercise diagnostics → `KILL`) stripped to birth-only; verified live on all three
+architectures — `BIRTH: Hermes live`/`BIRTH: Artemis live`, zero `KILL:` anywhere in any log,
+and an interactive `USE` round-trip confirmed Hermes genuinely reachable post-boot. Found and
+fixed a real regression along the way: Hermes's `CD-INIT` (arena init, and the thing that
+loads `lib.4th` into her own dictionary) was only ever invoked by the removed self-test —
+added an unconditional call at the end of her own `init.4th` so a real birth actually
+initializes her. Also added `STARTUP-BANNER` (scaffolding — prints `(no identity)` until
+`CERTVERIFY`/`MINT` exist to populate it from a thumbdrive's PKI fields), wired into each
+Tripod leg's own `init.4th`.
 
 **What this reshapes, not yet re-scoped:**
 - `WIREBIND` (§F.5) was deliberately left a breadcrumb, not fully designed, specifically