diff --git a/.github/workflows/python_lint.yml b/.github/workflows/python_lint.yml
index bd910c3..6b5370e 100644
--- a/.github/workflows/python_lint.yml
+++ b/.github/workflows/python_lint.yml
@@ -66,6 +66,14 @@ jobs:
         if: steps.changed-files-py.outputs.any_changed == 'true'
         run: python admin/scripts/check_report_local_paths.py
 
+      # The third element of an artifact's return tuple becomes the report's
+      # "located at" line and the LAVA manifest source_path, so it has to be real
+      # paths. Prose standing in for one points the examiner at a column that often
+      # holds a basename, and the location ends up nowhere in the report.
+      - name: Guard against prose returned as a source path
+        if: steps.changed-files-py.outputs.any_changed == 'true'
+        run: python admin/scripts/check_source_path.py
+
       # Fails only on warnings this pull request introduces. dleapp.py and
       # dleappGUI.py carry pre-existing warnings that are structural rather than
       # fixable -- wildcard imports are how those modules are put together -- so
diff --git a/admin/scripts/check_source_path.py b/admin/scripts/check_source_path.py
new file mode 100644
index 0000000..81ddb52
--- /dev/null
+++ b/admin/scripts/check_source_path.py
@@ -0,0 +1,180 @@
+"""Guard the third return element against prose standing in for a path.
+
+An artifact returns `(data_headers, data_list, source_path)`. The third element is
+not decoration. `artifact_processor` splits it on newlines, passes each piece
+through `Context.get_relative_path`, prints it as the report's
+
+    <artifact name> located at: <source_path>
+
+line, and writes it into the LAVA manifest as `source_path`. So it has to be real
+paths, newline joined.
+
+The defect this catches is a string constant standing in for one:
+
+    return data_headers, data_list, 'See source file(s) below'   # WRONG
+    return data_headers, data_list, 'Path column in the report'  # WRONG
+
+It reads as helpful and it is not. It points the examiner at a column that often
+holds a basename rather than a path, so the location of the evidence is nowhere in
+the report, and the LAVA manifest records prose where a consumer expects a path.
+
+The shape to write instead, accumulating after the function's own skip guards so a
+file that was matched but not parsed is not claimed as a source:
+
+    source_paths = set()
+    for file_found in files_found:
+        if <skip guard>:
+            continue
+        source_paths.add(str(file_found))
+        ...
+    return data_headers, data_list, '\\n'.join(sorted(source_paths))
+
+Pass the full staged path. The wrapper reduces it for you; that is the one place
+that reduction is done for the artifact.
+
+Two things this deliberately does NOT fail on:
+
+  * `''` on a branch that also returns an empty `data_list`. The wrapper writes no
+    report at all when there are no rows, so that string never reaches a report.
+    A "not found" message there is pointless but harmless, and `''` is correct.
+  * A variable holding a real path, however it was built. Only string literals are
+    reported, including a literal reached through a variable that is never assigned
+    anything else.
+
+Found by sweeping all five cores in 2026-08: 54 sites. iLEAPP had already been
+swept once (PRs #2022 and #2024, 140 modules) and the class had come back in
+ALEAPP and RLEAPP, which is why it is now a check rather than a sweep.
+
+Usage:
+  check_source_path.py [--root REPO_ROOT]
+
+Exits 1 when anything is found, 0 otherwise.
+"""
+
+import argparse
+import ast
+import os
+import sys
+
+DECORATORS = {'artifact_processor', 'artifact_processor_streaming'}
+
+STANDARD_NOTE = (
+    "Return the paths the function actually parsed, newline joined: "
+    "'\\n'.join(sorted(source_paths)). The wrapper makes them extraction relative."
+)
+
+
+def decorator_names(node):
+    names = []
+    for dec in node.decorator_list:
+        if isinstance(dec, ast.Name):
+            names.append(dec.id)
+        elif isinstance(dec, ast.Attribute):
+            names.append(dec.attr)
+        elif isinstance(dec, ast.Call):
+            func = dec.func
+            names.append(func.id if isinstance(func, ast.Name)
+                         else getattr(func, 'attr', ''))
+    return names
+
+
+def returns_no_rows(node):
+    """Whether this return also hands back an empty data_list literal."""
+    rows = node.value.elts[1]
+    return isinstance(rows, (ast.List, ast.Tuple)) and not rows.elts
+
+
+def scan_function(func):
+    """(line, text) for every literal masquerading as a source path."""
+    triples = [n for n in ast.walk(func)
+               if isinstance(n, ast.Return)
+               and isinstance(n.value, ast.Tuple)
+               and len(n.value.elts) == 3]
+    found = []
+    for ret in triples:
+        tail = ret.value.elts[2]
+
+        # Written straight into the return.
+        if isinstance(tail, ast.Constant) and isinstance(tail.value, str):
+            if tail.value and not returns_no_rows(ret):
+                found.append((ret.lineno, tail.value))
+            continue
+
+        # The quiet spelling: a name that is only ever assigned string literals.
+        if isinstance(tail, ast.Name):
+            assigns = [a for a in ast.walk(func) if isinstance(a, ast.Assign)
+                       and any(isinstance(t, ast.Name) and t.id == tail.id
+                               for t in a.targets)]
+            if not assigns:
+                continue
+            values = [a.value for a in assigns]
+            if not all(isinstance(v, ast.Constant) and isinstance(v.value, str)
+                       for v in values):
+                continue
+            for text in sorted({v.value for v in values}):
+                if text:
+                    found.append((ret.lineno, text))
+    return found
+
+
+def scan_module(path):
+    module = os.path.basename(path)
+    try:
+        with open(path, encoding='utf-8', errors='replace') as handle:
+            tree = ast.parse(handle.read())
+    except SyntaxError as err:
+        return [], f'{module}: could not parse ({err})'
+    violations = []
+    for node in ast.walk(tree):
+        if not isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)):
+            continue
+        if not DECORATORS.intersection(decorator_names(node)):
+            continue
+        for line, text in scan_function(node):
+            violations.append((module, node.name, line, text))
+    return violations, None
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    parser.add_argument('--root', default=None, help='repository root')
+    args = parser.parse_args()
+
+    root = args.root or os.path.dirname(
+        os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+    artifacts = os.path.join(root, 'scripts', 'artifacts')
+    if not os.path.isdir(artifacts):
+        print(f'No scripts/artifacts under {root}', file=sys.stderr)
+        return 2
+
+    violations, unreadable, modules = [], [], 0
+    for name in sorted(os.listdir(artifacts)):
+        if not name.endswith('.py'):
+            continue
+        modules += 1
+        found, problem = scan_module(os.path.join(artifacts, name))
+        violations.extend(found)
+        if problem:
+            unreadable.append(problem)
+
+    if violations:
+        print(f'Artifacts returning prose as source_path ({len(violations)}):')
+        for module, func, line, text in violations:
+            print(f'  {module}:{line}  {func}()  {text!r}')
+        print()
+        print(STANDARD_NOTE)
+        return 1
+
+    summary = (f'Checked {modules} artifact module(s): '
+               'no prose returned as a source path.')
+    if unreadable:
+        summary += f' {len(unreadable)} module(s) NOT checked.'
+    print(summary)
+    for problem in unreadable:
+        print(f'  {problem}')
+    return 0
+
+
+if __name__ == '__main__':
+    sys.exit(main())
diff --git a/admin/test/scripts/test_check_source_path.py b/admin/test/scripts/test_check_source_path.py
new file mode 100644
index 0000000..cbcb500
--- /dev/null
+++ b/admin/test/scripts/test_check_source_path.py
@@ -0,0 +1,111 @@
+"""Prove the source_path checker still detects the defects it exists to detect.
+
+check_source_path.py fails when an artifact returns a string constant where the
+report expects real paths. The class had already been swept out of iLEAPP once
+(PRs #2022 and #2024) and had come back in two sibling cores by the time it was
+made a check, so the check is the thing that has to keep working.
+
+The negative cases matter as much as the positive ones. A check wired into CI that
+fails correct code gets switched off, so the shapes that must stay silent are
+pinned here too: an empty string on a branch that returns no rows never reaches a
+report, and a variable holding a real path is not a literal.
+"""
+import importlib.util
+import pathlib
+import sys
+import tempfile
+import textwrap
+import unittest
+
+REPO_ROOT = pathlib.Path(__file__).resolve().parents[3]
+_MODULE_PATH = REPO_ROOT / 'admin' / 'scripts' / 'check_source_path.py'
+
+# admin/scripts is not a package, so load the module from its path.
+_spec = importlib.util.spec_from_file_location('check_source_path', _MODULE_PATH)
+csp = importlib.util.module_from_spec(_spec)
+sys.modules['check_source_path'] = csp
+_spec.loader.exec_module(csp)
+
+
+def findings_for(source):
+    with tempfile.TemporaryDirectory() as folder:
+        path = pathlib.Path(folder) / 'sample.py'
+        path.write_text(textwrap.dedent(source), encoding='utf-8')
+        violations, problem = csp.scan_module(str(path))
+    if problem:
+        raise AssertionError(problem)
+    return violations
+
+
+class Prose(unittest.TestCase):
+    def test_flags_prose_in_the_return(self):
+        found = findings_for('''
+            @artifact_processor
+            def demo(context):
+                data_list = [1]
+                return (), data_list, 'See source file(s) below'
+        ''')
+        self.assertEqual(len(found), 1)
+        self.assertEqual(found[0][3], 'See source file(s) below')
+
+    def test_flags_prose_reached_through_a_variable(self):
+        """appGrouplisting's spelling: assigned once, then returned."""
+        found = findings_for('''
+            @artifact_processor
+            def demo(context):
+                source_path = 'Path column in the report'
+                data_list = [1]
+                return (), data_list, source_path
+        ''')
+        self.assertEqual(len(found), 1)
+
+    def test_accepts_joined_real_paths(self):
+        self.assertEqual(findings_for('''
+            @artifact_processor
+            def demo(context):
+                source_paths = set()
+                data_list = []
+                for file_found in context.get_files_found():
+                    source_paths.add(str(file_found))
+                return (), data_list, '\\n'.join(sorted(source_paths))
+        '''), [])
+
+
+class MustStaySilent(unittest.TestCase):
+    def test_empty_string_on_a_no_rows_branch_is_fine(self):
+        """The wrapper writes no report when data_list is empty."""
+        self.assertEqual(findings_for('''
+            @artifact_processor
+            def demo(context):
+                if not context.get_files_found():
+                    return (), [], ''
+                return (), [1], '\\n'.join(sorted(paths))
+        '''), [])
+
+    def test_an_undecorated_helper_is_not_checked(self):
+        self.assertEqual(findings_for('''
+            def _helper(files_found):
+                return (), [1], 'See source file(s) below'
+        '''), [])
+
+    def test_a_variable_built_from_paths_is_not_a_literal(self):
+        self.assertEqual(findings_for('''
+            @artifact_processor
+            def demo(context):
+                source_path = get_file_path(context.get_files_found(), 'x.db')
+                return (), [1], source_path
+        '''), [])
+
+
+class TheRepoItself(unittest.TestCase):
+    def test_no_artifact_in_this_repo_returns_prose(self):
+        artifacts = REPO_ROOT / 'scripts' / 'artifacts'
+        offenders = []
+        for module in sorted(artifacts.glob('*.py')):
+            violations, _ = csp.scan_module(str(module))
+            offenders.extend(violations)
+        self.assertEqual(offenders, [], f'{len(offenders)} artifact(s) return prose')
+
+
+if __name__ == '__main__':
+    unittest.main()