diff --git a/src/echo_memory/server.py b/src/echo_memory/server.py
index b97a1e8..90bb0af 100644
--- a/src/echo_memory/server.py
+++ b/src/echo_memory/server.py
@@ -244,7 +244,6 @@ def record_recall_save(
     scope: str,
     fact_id: str,
     note: str,
-    recalled_by: str | None = None,
 ) -> dict:
     """Record that a fact you recalled from memory saved the user from
     re-explaining something to you.
@@ -254,21 +253,15 @@ def record_recall_save(
     answered something the user would otherwise have had to tell you again,
     and the fact was originally written by a DIFFERENT tool or a past session.
 
-    That last part is the whole point, and it is why this takes `fact_id`
-    rather than a `written_by` string. Pass the `fact_id` of the fact that
-    helped - every query_memory result carries one. The server reads that
-    edge's own `agent_id` and uses it as `written_by`; the caller does not get
-    to assert who wrote a fact.
+    That is why this takes only a `fact_id`. Pass the one carried by the
+    query_memory result that helped; the server reads that edge's own agent_id
+    for the writer, and uses its own configured agent id for the reader. You
+    assert neither. Both used to be caller-supplied, and each in turn let the
+    model being graded type its own evidence.
 
-    Until 2026-08-29 `written_by` was free text supplied by the caller. Nothing
-    checked the fact existed, so the number gating v1a was a string typed by
-    the model being graded. A fact_id is checkable, so the reading is
-    admissible.
-
-    recalled_by is you, defaulting to this server's own agent id. If the fact's
-    author and you are the same tool, the save is still recorded but does not
-    count toward the trial's bar - recalling your own note from ten minutes ago
-    is not the thing being measured.
+    If the fact's author and you are the same tool, the save is recorded but
+    does not count - recalling your own note from ten minutes ago is not the
+    thing being measured.
 
     note should be one sentence naming what it saved re-explaining, written so
     it still makes sense read cold in six months. Recording the identical note
@@ -282,7 +275,10 @@ def record_recall_save(
     except ConfigError as e:
         return {"error": str(e)}
 
-    recalled_by = recalled_by or _state.config.agent_id
+    # Not a parameter: an agent asserting which tool it is, to a criterion that
+    # measures whether two tools are involved, is the same fault the written_by
+    # fix closed on 2026-08-29.
+    recalled_by = _state.config.agent_id
     try:
         with _state.pool.connection() as conn:
             written_by = _author_of(conn, group_id, fact_id)
@@ -321,6 +317,12 @@ def record_recall_save(
         "recorded": True,
         "observation_id": recorded["id"],
         "already_recorded": not recorded["created"],
+        # Always returned, both branches. The caller supplies neither of these
+        # now, so the only way it can see what the server concluded about
+        # authorship - and check the save it just logged says what it meant -
+        # is for the response to state it.
+        "written_by": written_by,
+        "recalled_by": recalled_by,
         "counts_toward_gate": cross_tool,
         "cross_tool_saves": counts["cross_tool_saves"],
         "required": _observations.REQUIRED_SAVES,
diff --git a/tests/integration/test_recall_save_attribution.py b/tests/integration/test_recall_save_attribution.py
new file mode 100644
index 0000000..a1e49f8
--- /dev/null
+++ b/tests/integration/test_recall_save_attribution.py
@@ -0,0 +1,81 @@
+"""Neither side of a cross-tool save may be asserted by the caller.
+
+Criterion 6 counts a save only when the fact's author and its reader are
+different tools, and the agent supplying the evidence is the agent being
+graded. Both halves therefore have to come from somewhere the agent cannot
+choose.
+
+The writer half was closed on 2026-08-29: `written_by` used to be free text,
+so the number gating v1a was a string typed by the model being measured. It is
+now read from the fact's own edge via fact_id.
+
+The reader half stayed open until end-to-end testing on 2026-09-03 passed
+recalled_by="codex" from a claude-code server and watched it produce a counted
+cross-tool save. The reader is now the server's own configured agent id, which
+comes from the config the client launched it with, and is not a parameter."""
+
+import inspect
+
+import pytest
+from fake_embedder import REFERENCE, VectorEmbedder
+
+from echo_memory import server
+from echo_memory.infra.config import Config
+
+FACT = "a fact written by one tool and read by another"
+
+
+def _serve(migrated_db, agent_id):
+    config = Config(
+        user_id="ayush", agent_id=agent_id, database_url=migrated_db, project="echo-mem"
+    )
+    server.startup(config=config, embedder=VectorEmbedder({"probe": REFERENCE, FACT: REFERENCE}))
+    return config
+
+
+def _write(session_id):
+    result = server.write_episode(
+        "shared", session_id,
+        [{"name": "probe", "type": "test"}],
+        [{"source": "probe", "target": "probe", "relation_type": "is",
+          "fact": FACT, "confidence": "extracted"}],
+        entity_resolutions={"probe": {"resolved_to": "new"}},
+    )
+    return result["edges_created"][0]
+
+
+def test_the_reader_cannot_be_claimed(migrated_db):
+    """The exact call that used to work: one tool asserting it is another."""
+    _serve(migrated_db, "claude-code")
+    with pytest.raises(TypeError):
+        server.record_recall_save("shared", "1", "note", recalled_by="codex")
+
+
+def test_recalled_by_is_not_part_of_the_tool_contract(migrated_db):
+    _serve(migrated_db, "claude-code")
+    assert "recalled_by" not in inspect.signature(server.record_recall_save).parameters
+
+
+def test_one_tool_alone_cannot_produce_a_cross_tool_save(migrated_db):
+    _serve(migrated_db, "claude-code")
+    fact_id = _write("cc-1")
+
+    saved = server.record_recall_save("shared", str(fact_id), "read my own note back")
+
+    assert saved["recorded"] is True
+    assert saved["counts_toward_gate"] is False
+    assert saved["cross_tool_saves"] == 0
+
+
+def test_a_genuine_cross_tool_save_still_counts(migrated_db):
+    """Written by one server, read by another, exactly as two clients run."""
+    _serve(migrated_db, "codex")
+    fact_id = _write("codex-1")
+
+    _serve(migrated_db, "claude-code")
+    saved = server.record_recall_save("shared", str(fact_id), "codex knew this already")
+
+    assert saved["counts_toward_gate"] is True
+    assert saved["cross_tool_saves"] == 1
+    assert saved["written_by"] == "codex"
+    assert saved["recalled_by"] == "claude-code"