State only what a pending promote row proves

The delta review caught the new pending sentence asserting a cause: 'if the
row is still pending minutes later, the event did not land'. A promote that
prepares a source stays pending, untouched, for as long as the clone runs (up
to CLONE_TIMEOUT_SEC = 900 s), so a model following that sentence would mint a
second id while the first was alive, which is the duplicate-root class this
package cures. The sentence now gives the two times, the longest honest wait
and the fact that a new promote is a new id, and leaves the judgement to the
model.

ouroboros/tools/control_task_results.py

@@ -197,6 +197,7 @@ def _get_task_result(
             status="unavailable", code="LEGACY_UNAVAILABLE",
             text=f"Task {task_id}: unknown or not yet registered",
         ))
+    from ouroboros.project_sources import CLONE_TIMEOUT_SEC
     from ouroboros.routing_wait import is_emitted_admission_stub
 
     if is_emitted_admission_stub(data):
@@ -210,9 +211,10 @@ def _get_task_result(
             text=(
                 f"Task {task_id}: admission pending since {since} (now {utc_now_iso()}) - the promote "
                 "was emitted and this row carries no supervisor receipt yet, neither scheduled nor "
-                f"refused. Read get_task_result({task_id}) again before promoting the same work: the "
-                "supervisor answers within seconds of draining its queue. If the row is still "
-                "pending minutes later, the event did not land and a new promote is the way forward."
+                f"refused. Read get_task_result({task_id}) again before promoting the same work: a "
+                f"promote that prepares a source can stay pending for up to {CLONE_TIMEOUT_SEC // 60} "
+                "minutes. A new promote is a NEW task id; this row does not forbid one, and whether "
+                "this one is lost is yours to judge from the two times above."
             ),
         ))
     if bool(include_authority) or bool(include_work_order_source) or bool(include_completion_source):

tests/test_promote_admission_pending.py

@@ -309,8 +309,9 @@ def test_the_attachment_refusal_without_a_stub_still_writes_no_result(swarm_host
 
 def test_the_pending_sentence_never_forbids_a_new_promote_for_ever(tmp_path, _short_confirmation_window):
     """The row proves one thing: emitted, no receipt yet. An event can be lost (a
-    supervisor restart drops its in-memory queue), so the sentence hands the
-    decision back with the facts instead of a standing ban."""
+    supervisor restart drops its in-memory queue) and a source can take minutes to
+    prepare, and the row cannot tell the two apart; so the sentence gives the facts
+    and hands the decision back, with neither a standing ban nor an invented cause."""
     from ouroboros.tools.control_routing import _promote_chat_to_task
     from ouroboros.tools.control_task_results import _get_task_result
 
@@ -321,4 +322,7 @@ def test_the_pending_sentence_never_forbids_a_new_promote_for_ever(tmp_path, _sh
     read = _get_task_result(ctx, task_id)
     assert "admission pending since" in read and f"get_task_result({task_id})" in read
     assert "do not promote the same work a second time" not in read
-    assert "did not land" in read and "new promote" in read
+    # Facts only: the row cannot tell a lost event from a source still being prepared,
+    # so the sentence names the longest honest wait and leaves the judgement to the model.
+    assert "did not land" not in read and "within seconds" not in read
+    assert "can stay pending for up to 15 minutes" in read and "NEW task id" in read