diff --git a/tests/commands/test_scope.py b/tests/commands/test_scope.py
index 9b09042..e80e344 100644
--- a/tests/commands/test_scope.py
+++ b/tests/commands/test_scope.py
@@ -269,14 +269,22 @@ def test_attest_writes_none_when_no_current_credential_resolves(
     db_path: Path, valid_config_path: Path, monkeypatch: pytest.MonkeyPatch
 ) -> None:
     _at_a_terminal(monkeypatch, yes=False)
+    # STATED, not inherited from the machine (#742). This asserted that the real
+    # `current_credential_fingerprint` resolves to None, which is true only when the process
+    # environment, the `.env` at `default_env_path()` and the OS keychain are ALL empty -- true
+    # in CI, false for any contributor with a configured deployment, where `CDP_API_KEY`
+    # resolves from the `.env` and this failed. The behaviour under test is what the command
+    # WRITES when nothing resolves, so the "nothing resolves" half belongs in the fixture.
+    monkeypatch.setattr(
+        "keel.commands.scope.current_credential_fingerprint", lambda venue: None
+    )
     result = _run(
         db_path, valid_config_path, "scope", "attest", "--read-only", "--venue", "coinbase",
     )
     assert result.exit_code == 0, result.output
     record = _repo_at(db_path).get_venue_trade_scope("coinbase")
     assert record is not None
-    # No credentials are configured in this test's environment, so the real
-    # `current_credential_fingerprint` resolves to None -- written as-is, not defaulted away.
+    # An unresolvable credential is written AS None, not defaulted away.
     assert record.credential_fingerprint is None
 
 
diff --git a/tests/execution/test_executor.py b/tests/execution/test_executor.py
index c792c72..73f700f 100644
--- a/tests/execution/test_executor.py
+++ b/tests/execution/test_executor.py
@@ -3636,8 +3636,22 @@ def test_confirming_replaces_a_stale_fingerprint_rather_than_carrying_it_forward
         credential_fingerprint="stale" + "0" * 27,
     )
 
-    with mock.patch.object(
-        executor, "current_credential_fingerprint", return_value="fresh" + "0" * 27
+    # BOTH import sites, and they are patched to DIFFERENT values on purpose (#742).
+    # `current_credential_fingerprint` is imported independently by `execution/guards.py`
+    # (rail 20) and by `execution/executor.py`. This test is about what the EXECUTOR writes on
+    # confirmation, so rail 20 must not veto first -- and it vetoes on a resolvable fingerprint
+    # that disagrees with the stored "stale000...". `None` is exactly the "no evidence to
+    # disagree with" case that lets the entry through to the confirm step.
+    #
+    # That None was previously supplied by the machine rather than the test: CI has no
+    # credentials, so the unpatched `guards` call returned None and the test passed. On a
+    # contributor's box with a `.env`, it returned a real fingerprint, rail 20 vetoed, `execute`
+    # never reached the confirm, and this assertion failed with the fingerprint still stale.
+    with (
+        mock.patch.object(
+            executor, "current_credential_fingerprint", return_value="fresh" + "0" * 27
+        ),
+        mock.patch.object(guards, "current_credential_fingerprint", return_value=None),
     ):
         execute(_enter_signal(), FakeBroker(), repo, _config(), mode="autonomous", now_ts=NOW_TS)