docs: decisions 133 (redeploy without rows 91 and 93) and 134 (Filbert to Opus 5.5), P19 #250 close record

Co-Authored-By: Claude Opus 5.5 <[email protected]>

docs/plans/2026-09-26_lead-decisions.md

@@ -2966,3 +2966,47 @@ which stay with him. Each item names who decided it and what happened.
       It takes the test slot ahead of rows 80 and 88.
     - The bus restart doesn't move earlier for this. Decision 118's
       order stands.
+
+133. **The redeploy ships without rows 91 and 93 (2026-10-13, Mos's
+    ruling after Jason rebooted dragon-lin at 13:40:09Z).** The bus unit
+    stayed down after the reboot because it was never enabled. Decision
+    118's redeploy went ahead instead of waiting for the sandbox rows.
+    - Sage reinstalled the unit with `scripts/bus-service.sh install`
+      from canonical 8ac8055c. The only change is the
+      `EnvironmentFile=-%h/.config/mosaic-dev/secrets/claude-session.env`
+      line. Mos's pm.conf drop-in still adds `--pm`.
+    - Sage started the bus at 14:04:18Z. The unit is active with
+      NRestarts 0, /healthz on :7332 returns ok, and `launches list`
+      shows PM run r71a2676a33a8. The bus runs the code from rows 89,
+      90, 92 and 94.
+    - Install doesn't enable the unit and Sage didn't. Mos enabled it at
+      14:09:15Z without restarting it. Undo with `systemctl --user
+      disable mosaic-bus@mosaic-stack.service`. Whether install should
+      enable units is a later row, not urgent.
+    - Row 99 is dropped: the fallback only helps against a pre-94 bus,
+      which only a rollback could bring back. Queue rev 907 blocks it
+      with the reason (no dropped state exists, and parking needs Jason
+      from briefed). It stays blocked until Jason is next in the queue.
+      Gitea #1593 is closed (comment 28146) and Vikunja P19 #250 is done
+      (comment 134).
+    - Rows 91 and 93 need their own bus restart when they land. Sage
+      asks Mos for it then.
+    - Next: Jason's PM re-test (#228, #232). Row 85 waits on it.
+
+134. **Filbert moves to Claude Opus 5.5 (2026-10-13, a seat change on
+    Jason's direct input).** Jason reported that OpenAI's cyber-risk
+    filter keeps blocking Filbert's reviews on gpt-6.1-sol. The thread
+    it blocked was the retired 50342f6a; f712d8af approved row 97 round 3
+    at 13:18Z without a flag. The filter would still fire on any
+    security-adjacent review, so the seat moves while no Filbert review
+    is in flight.
+    - New thread 048f3c63 (Claude, claude-opus-5-5). Codex thread
+      f712d8af is retired; nobody messages it.
+    - Opus roster seats are now Darkwing, Dewey and Filbert, three of
+      decision 45's four. Sage's own thread isn't counted.
+    - Cost: no GPT reviewer remains, so every review is Claude on Claude.
+      Decision 122 chose Codex for reviewer diversity. Rocko stays on
+      Codex as a builder.
+    - Decision 130's split stands: Filbert reviews non-sandbox rows only,
+      and Dewey keeps rows 91 and 93.
+    - Mos had no objection.