diff --git a/docs/designs/server/compass-comms-multi-actor-test-coverage/design.md b/docs/designs/server/compass-comms-multi-actor-test-coverage/design.md
index 9d4377c7d..6befe556c 100644
--- a/docs/designs/server/compass-comms-multi-actor-test-coverage/design.md
+++ b/docs/designs/server/compass-comms-multi-actor-test-coverage/design.md
@@ -147,6 +147,13 @@ The offline-redelivery records (`docs/designs/server/compass-mention-offline-red
 func TestCommsOfflineRedeliveryOnSessionStart(t *testing.T) // //go:build podman
 ```
 
+**AMENDED post-freeze (RIG-3532, implemented in `legcomms_redeliver_test.go`).** Two of this task's prescriptions above are falsified by the code; both were verified against the tree before implementation, and the shipped test takes a different shape. The reasoning in the paragraph above — assert durably, never on the live fan — stands and is why the test reads the store.
+
+1. **The durable surface is `UndeliveredMessages`, not `owed_mentions`.** `CountOwedMentions`/`ClearOwedMention` are the wrong surface here: `owed_mentions` rows are written only by `RecordOwedMention`, whose callsites (`delivery/dispatch.go`, `routeMentionsFor` and `routeAskAnswerFor`) require the message to `@`-mention the agent **and** the agent to be outside the channel's sweep set. `InSweepSet` is satisfied by `cm.channel_id = aa.home_channel_id`, so a post to the agent's own home channel never records a row — `CountOwedMentions` reads 0 both before and after the sweep, and a test asserting on it would be green while proving nothing. `(*store.Store).UndeliveredMessages` is what the start-sweep itself consumes (`drainStarts` → `sweepSession`), so it is the surface that actually moves.
+2. **"Offline" must be a despawn, not a never-started agent.** "Post to a provisioned-but-not-started agent's home channel, then `StartSession`" is unreachable: the post itself starts the agent. `fanOut` resolves the agent as a recipient (same home-channel disjunct), finds no live session, calls `wake`, and `freshStart` resolves the placement `Provision` already recorded (`service.go` `RecordAgentPlacement`) and calls `hub.Start` — the behavior `TestWakeAgentNeverStartedFreshStarts` pins. The waker is wired in the real assembly the e2e fixture runs (`server/sinks.go` `SetAgentWaker`). A subsequent `StartSession` would be a second, racing start. The shipped test instead posts **after** `RemoveWorkspace` deletes the placement, which makes the wake a benign `no_placement` no-op (`wakeOnce`'s `errWakeNoPlacement` arm), so the message stays owed across a real lifetime boundary. It therefore costs two containers rather than the one estimated above, and drives one warm turn in lifetime 1 so the resume has a transcript to reconstruct.
+
+The test also asserts the message **is** owed before the resume, so the "no longer owed" post-condition cannot pass vacuously.
+
 ### T6 — DB tier: supervisor orchestration depth (`supervisor_orchestration_pgtest_test.go`)
 
 The sharpest thin spot: one test func today (`supervisor_orchestration_pgtest_test.go:38` `TestSupervisorAssignsToTwoWorkersAuditable` is the file's only `func Test`). Add: (a) **many concurrent conversations** — one supervisor, 4 workers, per-worker topics in one coordination channel; interleaved posts; assert per-topic `ListMessages` returns exactly that worker's thread in order (per-agent conversation isolation at the DB tier, complementing `agent_conversation_pgtest_test.go`); (b) worker→supervisor upward reports land on the worker's topic with author intact; (c) `SearchMessages` audit across 4 workers scoped by membership (extending the existing outsider-zero-hits case); (d) coordination-channel membership reflects a mid-scenario `ReparentAgent` (consumes `go/internal/comms/coordination.go:197-203` `ReconcileCoordinationMembership`).