decisions, build-log, sessions: cutover live, 109 landings are live Console changes

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

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

@@ -2413,3 +2413,27 @@ which stay with him. Each item names who decided it and what happened.
       brief is Sage's to write. M1 is decision 104. Under decision 96,
       when T3 Mos retires is still Jason's call after the M3 trial, and
       nobody puts it to him before then.
+109. **Landings in the canonical checkout are live Console changes
+    (2026-10-12).** The cutover Console (decision 103, started by Mos
+    at 17:21:41Z after Jason's go at 17:21:09Z) runs from
+    `/mnt/storage/src/mosaic-stack`, the checkout Sage lands rows into.
+    Its webui modules load once at start. But every read and send spawns
+    a fresh `node packages/bus/src/human-cli.mjs` from the same tree
+    (`packages/webui/src/bus.mjs:31`). So a landing that touches
+    `packages/bus` changes the live Console on its next request.
+    `Restart=on-failure` loads whatever webui and cli code the tree
+    holds at that moment.
+    - Until the Console runs from a pinned copy, Sage treats a landing
+      that touches `packages/webui`, `packages/bus` or `packages/cli`
+      as a live change. Sage tells Mos before it lands, gives the landed
+      commit after, and runs `scripts/webui-service.sh status` and
+      /healthz. The landing doesn't restart the unit. A webui change
+      reaches the Console when Jason or Mos restarts it.
+    - The landing itself stays the gate recipe: `git apply --index`
+      then `sha256sum -c` in one step. Seats don't edit or check out
+      branches in the canonical checkout, which is already the rule.
+    - Follow-up row for 2026-10-13: run the Console from a pinned
+      worktree or release copy at a landed commit, so landings stop
+      being deploys. Pointing the unit at it is a reinstall, which is
+      Mos's step under decisions 102 and 103. It goes in the same window
+      as decision 101's uid reinstall.