diff --git a/proposals/4140-delayed-events-futures.md b/proposals/4140-delayed-events-futures.md
index c9f5c41b1..d337bc05f 100644
--- a/proposals/4140-delayed-events-futures.md
+++ b/proposals/4140-delayed-events-futures.md
@@ -187,10 +187,16 @@ Content-Type: application/json
 Where the `action` is `send`, the homeserver **should** apply rate limiting to provide mitigation against the
 [High Volume of Messages](https://spec.matrix.org/v1.11/appendices/#threat-high-volume-of-messages) threat.
 
-Once a delayed event has been sent or canceled, it is finalized and cannot be
-accessed via this endpoint.
+The server will return a `M_NOT_FOUND` error if the `delay_id` is not recognized or was already canceled.
 
-The server will return a `M_NOT_FOUND` error if the `delay_id` is not recognized.
+The success response body is always an empty object. A future MSC may define additional keys, such as returning
+the event ID for `send` or the new expected send time for `restart`.
+
+To allow safely retrying requests, the `send` action should return HTTP 200 even if the event was already sent
+successfully by a previous request or by the delay timing out.
+
+If sending the event via a `send` call fails, or if the event already failed previously, the endpoint returns the error
+that happened while sending the event (e.g. `M_FORBIDDEN` if the user doesn't have permission to send the event).
 
 The primary point of rate limiting is event sending when the delay times out or the event is sent using the `send`
 action. However, servers can choose to rate limit the the management endpoints themselves as well if necessary.