diff --git a/scripts/check-doc-count.py b/scripts/check-doc-count.py
index 897d3fe2..0a0c9068 100644
--- a/scripts/check-doc-count.py
+++ b/scripts/check-doc-count.py
@@ -108,6 +108,16 @@ def _resolve_root() -> Path:
 # the reader straight into a second failure. A hint is only worth printing if it
 # runs; keep the spelling here and let tests/test_check_doc_count.py prove the
 # other sites agree with it.
+#
+# Scope of that measurement, added 2026-09-13 (cyc20260913-122923) after walking
+# into it: "this form exits 0" is a property of a **synced** checkout. The
+# measurement above was taken in the main clone, which is synced. In a fresh
+# worktree the same spelling exits 2 without measuring anything, because
+# `uv run --no-sync` has created an empty `.venv` there and both `python` and
+# `python3` resolve to it - so the spelling is necessary but not sufficient, and
+# the hint it appears in must not be printed when the environment, not the
+# interpreter choice, is what is missing. See the pytest-missing branch in
+# `measured_count`.
 INVOCATION = "uv run --no-sync python3 scripts/check-doc-count.py"
 
 # The stored form, as it appeared in Agent.md: the command followed by the count
@@ -185,9 +195,31 @@ def measured_count() -> int:
             "(its output could not be decoded)"
         )
     if proc.returncode != 0:
+        detail = (proc.stdout[-2000:] + proc.stderr[-2000:]).strip()
+        # Two causes, two remedies - and the second one is *not* "use the right
+        # interpreter". Measured 2026-09-13 (cyc20260913-122923) in a fresh
+        # review worktree: `uv run --no-sync` there had produced an empty `.venv`
+        # (`site-packages` holding only `_virtualenv.pth` and `_virtualenv.py`),
+        # and `python` and `python3` both resolve to it - so the invocation this
+        # tool prints as its own remedy failed with byte-identical output, rc 2,
+        # sending the reader in a circle. Reading that output as a wrong
+        # interpreter is the misdiagnosis: there is no interpreter in this
+        # checkout that has pytest, so advising a different one cannot help.
+        if "No module named pytest" in detail:
+            raise DocCountError(
+                "pytest is not installed in the interpreter running this tool "
+                f"({sys.executable}), so no test was collected:\n"
+                + detail
+                + "\n\nThis is an unsynced checkout, not a wrong-interpreter "
+                "problem - a fresh worktree or clone gets an empty `.venv`, and "
+                "`python` and `python3` both resolve to it, so re-running "
+                f"`{INVOCATION}` here fails identically. Run `uv sync` in this "
+                "checkout first, or run this tool from a checkout whose "
+                "environment is already synced."
+            )
         raise DocCountError(
             f"pytest --collect-only failed (rc={proc.returncode}):\n"
-            + (proc.stdout[-2000:] + proc.stderr[-2000:]).strip()
+            + detail
             + "\n\nhint: run this with the project interpreter, e.g."
             f" `{INVOCATION} --measure`"
         )
diff --git a/tests/test_check_doc_count.py b/tests/test_check_doc_count.py
index 58c2e6d1..ca3f36d4 100644
--- a/tests/test_check_doc_count.py
+++ b/tests/test_check_doc_count.py
@@ -528,6 +528,56 @@ class _Proc:
     assert f"`{mod.INVOCATION} --measure`" in str(excinfo.value)
 
 
+def test_a_missing_pytest_is_diagnosed_as_an_unsynced_checkout(monkeypatch) -> None:
+    """The remedy must not be the command that just failed.
+
+    Measured state this pins (2026-09-13, `cyc20260913-122923`): in a fresh
+    review worktree the tool printed its own `INVOCATION` as the fix, and running
+    that spelling produced byte-identical output, rc 2 - `uv run --no-sync` had
+    left an empty `.venv` there and both `python` and `python3` resolve to it, so
+    "use the project interpreter" is a circle. The message must name the
+    environment instead.
+    """
+    fresh = _load_module()
+
+    class _Proc:
+        returncode = 1
+        stdout = ""
+        stderr = "/some/checkout/.venv/bin/python3: No module named pytest\n"
+
+    monkeypatch.setattr(fresh.subprocess, "run", lambda *a, **k: _Proc())
+    with pytest.raises(fresh.DocCountError) as excinfo:
+        fresh.measured_count()
+    message = str(excinfo.value)
+    assert "No module named pytest" in message
+    assert "/some/checkout/.venv/bin/python3" in message, "the child's own words"
+    assert "unsynced" in message, "the cause, named"
+    assert "uv sync" in message, "a remedy that can actually work here"
+
+
+def test_a_real_collection_failure_keeps_the_invocation_hint(monkeypatch) -> None:
+    """The other cause of the same non-zero exit: pytest ran, and it failed.
+
+    Without this arm, treating every non-zero collection as an unsynced checkout
+    would be green - which would take the invocation hint away from the case it
+    was written for.
+    """
+    fresh = _load_module()
+
+    class _Proc:
+        returncode = 2
+        stdout = "ERROR: file or directory not found: tests/\n"
+        stderr = ""
+
+    monkeypatch.setattr(fresh.subprocess, "run", lambda *a, **k: _Proc())
+    with pytest.raises(fresh.DocCountError) as excinfo:
+        fresh.measured_count()
+    message = str(excinfo.value)
+    assert f"`{fresh.INVOCATION} --measure`" in message
+    assert "uv sync" not in message
+    assert "unsynced" not in message
+
+
 # --- which tree was scanned --------------------------------------------------