diff --git a/src/greatminds/data/roles/DEVELOPER.md b/src/greatminds/data/roles/DEVELOPER.md
index c1d2718..0056bed 100644
--- a/src/greatminds/data/roles/DEVELOPER.md
+++ b/src/greatminds/data/roles/DEVELOPER.md
@@ -21,6 +21,25 @@ DEVELOPER implements backend/product-code tasks planned by ARCHITECT-PLANNER.
 5. For `plan.stand_required: true`, files a stand request with
    `greatminds stand request ... --evidence-for <task-id>`.
 6. Records any stand request id/result/caveat in the implementation block.
+
+   **If a prior iter put the stand in `down`** because of a
+   playbook bug your iter-N fix resolves: after the fix is in
+   your worktree, send STAND-KEEPER a wake with the canonical
+   `FIX-LANDED` body shape (SK reads the body verbatim and will
+   then run `stand up` on its next tick — `greatminds stand up`
+   is SK-only):
+
+   ```bash
+   greatminds inbox send STAND-KEEPER --kind wake \
+     --task <task-id> \
+     --body "FIX-LANDED for <task-id> stand-profile <profile>. \
+       Worktree .worktrees/<task-id>/coordination/stand-profiles/<profile>.yaml \
+       carries iter-<N> changes: <one-line summary>. Please \
+       greatminds stand up --reason 'iter-<N> fix landed for <task-id>'."
+   ```
+
+   Don't ask MAINTAINER to clear the stand — `stand up` is gated
+   to STAND-KEEPER and MAINTAINER has no override.
 7. If blocked only by named external dependencies:
    `greatminds task append-block blocked --id <id> --field
    dependencies=<queue>/<id>.yaml,... --field resume_to=feature_dev`
diff --git a/src/greatminds/data/roles/STAND-KEEPER.md b/src/greatminds/data/roles/STAND-KEEPER.md
index 362fd18..8daddff 100644
--- a/src/greatminds/data/roles/STAND-KEEPER.md
+++ b/src/greatminds/data/roles/STAND-KEEPER.md
@@ -89,8 +89,30 @@ gone. The singleton stand is driven by `coordination/.stand/state.yaml`.
    `preparing` lease on a later tick.
 5. **On deploy or infra failure:** run
    `greatminds stand down --reason "profile failed: <step>"`, naming
-   the failed profile step. Queue processing pauses.
-   After recovery, run `greatminds stand up --reason "<note>"`.
+   the failed profile step. Queue processing pauses until you run
+   `greatminds stand up`. While the stand is `down`, **do not act
+   on bare wake pings or wakes that describe a lease** — there is
+   no active lease to deploy, and re-running an unfixed playbook
+   would just re-fail.
+
+   **Clearing the `down` state.** Run `greatminds stand up --reason
+   "<note>"` only after receiving an inbox message whose body
+   carries an explicit *fix-landed* signal from the implementer
+   (DEVELOPER) or the holder (TESTER / EXPLORER). The canonical
+   body shape is:
+
+   > `FIX-LANDED for <task-id> stand-profile <profile>. Worktree
+   > .worktrees/<task-id>/coordination/stand-profiles/<profile>.yaml
+   > carries iter-<N> changes: <one-line summary>. Please
+   > greatminds stand up --reason "iter-<N> fix landed for <task-id>".`
+
+   Verify the worktree carries the named change (a quick `grep`
+   or `diff` is enough — you do not need to re-deploy yet), then
+   transition with `stand up`. If the body lacks the `FIX-LANDED`
+   prefix or doesn't name a worktree path you can verify, ignore
+   the message — the stand stays `down` until a properly-shaped
+   signal arrives. This intentional friction is what prevents a
+   stray wake from re-running unfixed code.
 6. **Never release the holder's active lease.** TESTER or EXPLORER
    runs `greatminds stand release --lease-id <lease_id> --result
    pass|fail|partial` after its own probes.
diff --git a/src/greatminds/data/roles/TESTER.md b/src/greatminds/data/roles/TESTER.md
index 535a2d1..407babc 100644
--- a/src/greatminds/data/roles/TESTER.md
+++ b/src/greatminds/data/roles/TESTER.md
@@ -51,6 +51,27 @@ tests block.
    only in the product task's `tests` block. The old
    `stand_requests/` -> `stand_wip/` -> `stand_done/` path is removed
    from the active workflow.
+
+   **When the stand sits in `down`** (e.g. a prior deploy hit a
+   playbook bug and you've bounced the task to DEVELOPER for a fix):
+   wait for DEV's inbox info that iter-N has landed. Then, if you
+   want to drive the recovery yourself, send STAND-KEEPER a wake
+   with the canonical `FIX-LANDED` body shape:
+
+   ```bash
+   greatminds inbox send STAND-KEEPER --kind wake \
+     --task <task-id> \
+     --body "FIX-LANDED for <task-id> stand-profile <profile>. \
+       Worktree .worktrees/<task-id>/coordination/stand-profiles/<profile>.yaml \
+       carries iter-<N> changes: <one-line summary>. Please \
+       greatminds stand up --reason 'iter-<N> fix landed for <task-id>'."
+   ```
+
+   `greatminds stand up` is SK-only — TESTER cannot call it
+   directly. The wake's `FIX-LANDED` body is what SK reads to know
+   the state is safe to clear. Bare wake pings or wakes that talk
+   about an active lease are not enough; SK will (correctly) ignore
+   them while the stand is `down`.
 6. For `plan.stand_required: true`:
    - waits for its lease to become ready via SK's inbox-info,
    - probes the stand directly and records lease-based stand evidence,