docs: lead decision 69, Rocko moves to a Claude thread under R26 (sage)

Jason's R26 via Mos: Claude first, Codex only on gpt-6.1-sol, never
Astra. Rocko's Codex thread ran gpt-6-astra; the seat now runs as T3
thread b84bb264 on Opus 5.5, because its only row, S4, builds mosaic
decide and the gated decision DM. Roster updated. Filbert's launcher
still pins Astra; recorded as a follow-up, unused while Filbert runs in T3.

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

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

@@ -1251,3 +1251,33 @@ which stay with him. Each item names who decided it and what happened.
       2.7.0 OpenAPI), not a bare array. The probe's `items ?? body`
       read hid it. The runbook's listing line now reads `.items[]?`,
       which also covers a null `items`.
+69. **R26 on the stack: Rocko moves to a Claude thread (2026-10-08).**
+    Source: Mos, thread 1eba59e5, relaying Jason's R26. Agentic work
+    runs on Claude models first. Codex sessions use `gpt-6.1-sol` and
+    never Astra, and T3's Codex harness defaults to `gpt-6-astra`.
+    - Stack seats in T3 on 2026-10-08: Sage, Darkwing, Dewey and
+      Filbert run Claude Opus 5.5. Researcher (9bac0794) runs Claude
+      Sonnet 5.5. Rocko (9b95add7) ran Codex on `gpt-6-astra`.
+    - Mos's option (a), a Claude thread for Rocko, chosen over (b), a
+      Codex thread on Sol. Reviews already pair Claude with Claude, and
+      a second model family would buy reviewer diversity, but R26 puts
+      Claude first, and (b) needs a reason I don't have.
+    - The new Rocko thread is b84bb264, on Opus 5.5, not the roster's
+      Sonnet. Rocko's only open row is S4, which builds
+      `mosaic decide` (it refuses inside an agent run) and the gated
+      decision DM to Jason. Under Mos's split, authority and security
+      work runs on Opus. T3 can't change a thread's model, so the
+      thread starts on the model S4 needs. This stays inside decision
+      45's limit of four Opus sessions. The roster in `agents/README.md`
+      records it.
+    - Clean point: the old thread was idle, row 39 is queued and not
+      briefed, and the earlier packets stay under `agents/rocko/work/`.
+      Nobody messages the old thread again, because that spends Astra
+      usage. Its rules carry over: no Gitea credentials, no launch, no
+      commits.
+    - Follow-up: `agents/filbert/launch.sh` pins
+      `openai-codex/gpt-6-astra:low`, and `scripts/test-darkwing-launch.mjs`
+      asserts it. Filbert runs on Claude in T3, so nothing uses that
+      launcher today. Its Astra pin conflicts with R26, and the roster
+      says not to use it until a reviewed change moves it to Claude or
+      Sol.