diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5e0b84c22..bd7bb85bc 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -17,6 +17,7 @@ Full release notes with details on each version: [GitHub Releases](https://githu
 - Fix: a PHP `use` import written with a leading-backslash / fully-qualified prefix now resolves to its target definition instead of being dropped (#2661, thanks @ousamabenyounes).
 - Fix: an unresolved local JS/TS import (to a file absent from the scan) now emits a stable, portable `ref` target id instead of leaking a per-checkout absolute-path slug (#2457, thanks @rohit-jsfreaky).
 - Fix: `graphify benchmark` no longer crashes on a node whose label is `None` (#2674, thanks @Arthuro0103).
+- Fix: a corrupt (invalid-JSON) semantic cache entry is no longer swallowed as a silent miss; `check_semantic_cache` now emits one aggregate warning naming how many entries failed to parse, so a cache that fails on every run (silently re-billing the extraction) is diagnosable instead of invisible (#2405, thanks @jorgefusterr).
 
 ## 0.9.40 (2026-08-11)
 
diff --git a/graphify/cache.py b/graphify/cache.py
index ea94227ab..bdb6cbd1e 100644
--- a/graphify/cache.py
+++ b/graphify/cache.py
@@ -80,6 +80,12 @@ def _cleanup_stale_ast_entries(ast_base: Path, current_dir: Path) -> None:
 # check_semantic_cache can report N to the user (#1939).
 _legacy_semantic_hits = 0
 
+# Count of cache entries that failed to parse as JSON this process. A corrupt
+# entry is not a miss: left in place it fails on every future run, silently
+# re-extracting (and, for semantic kinds, re-billing) the file forever. The
+# counter lets check_semantic_cache surface one aggregate warning (#2405).
+_corrupt_cache_entries = 0
+
 # Prompt-file fingerprints already computed, keyed by (path, size, mtime_ns) —
 # the same stat signature the hash index uses. check_semantic_cache resolves the
 # prompt once per FILE in the corpus, so without this a 500-doc run re-reads and
@@ -896,7 +902,7 @@ def load_cached(path: Path, root: Path = Path("."), kind: str = "ast",
     ``merge_existing``) pass allow_legacy=False.
     Returns None if no cache entry or file has changed.
     """
-    global _legacy_semantic_hits
+    global _legacy_semantic_hits, _corrupt_cache_entries
     location = cache_root if cache_root is not None else root
     try:
         h = file_hash(path, root, cache_root=cache_root)
@@ -912,7 +918,14 @@ def load_cached(path: Path, root: Path = Path("."), kind: str = "ast",
     if entry.exists():
         try:
             result = json.loads(entry.read_text(encoding="utf-8"))
-        except (json.JSONDecodeError, OSError):
+        except json.JSONDecodeError:
+            # Corrupt entry, not a miss: a truncated write or a bad producer
+            # (e.g. unescaped Windows backslashes in source_file) leaves JSON
+            # that fails to parse on every future run, so the file is silently
+            # re-extracted forever. Count it so the run can report it (#2405).
+            _corrupt_cache_entries += 1
+            return None
+        except OSError:
             return None
         # A ``partial`` entry was produced from a truncated LLM response and
         # covers only part of the file's symbols. Serving it as authoritative
@@ -1154,6 +1167,7 @@ def check_semantic_cache(
     cached_hyperedges: list[dict] = []
     uncached: list[str] = []
     legacy_before = _legacy_semantic_hits
+    corrupt_before = _corrupt_cache_entries
 
     for fpath in files:
         p = Path(fpath)
@@ -1180,6 +1194,18 @@ def check_semantic_cache(
             stacklevel=2,
         )
 
+    corrupt = _corrupt_cache_entries - corrupt_before
+    if corrupt:
+        warnings.warn(
+            f"{corrupt} semantic cache entr{'y' if corrupt == 1 else 'ies'} could "
+            "not be parsed as JSON and were treated as misses, so those files were "
+            "re-extracted. A corrupt entry stays on disk and fails again every run; "
+            "run with --force (or GRAPHIFY_FORCE=1) to rewrite them, or clear the "
+            "cache to stop paying for the re-extraction (#2405).",
+            RuntimeWarning,
+            stacklevel=2,
+        )
+
     return cached_nodes, cached_edges, cached_hyperedges, uncached
 
 
diff --git a/tests/test_cache.py b/tests/test_cache.py
index 510ca430d..67135b393 100644
--- a/tests/test_cache.py
+++ b/tests/test_cache.py
@@ -1455,3 +1455,34 @@ def test_file_hash_fastpath_still_serves_a_settled_file(tmp_path, monkeypatch):
     second = file_hash(f, tmp_path)
     assert second == first
     assert reads == [], "settled file was re-read; the stat fastpath is dead"
+
+
+def test_corrupt_semantic_entry_warns_and_is_a_miss(tmp_path):
+    """A corrupt (invalid-JSON) cache entry must not be silently swallowed
+    (#2405). Left unreported it fails to parse on every future run, re-billing
+    the semantic extraction forever with no diagnostic. check_semantic_cache
+    treats it as a miss (uncached) AND emits one aggregate warning naming the
+    count, mirroring the pre-fingerprint legacy-hit warning."""
+    from graphify.cache import (
+        check_semantic_cache,
+        save_semantic_cache,
+        cache_dir,
+    )
+
+    f = tmp_path / "doc.md"
+    f.write_text("# Doc\n\nBody.\n")
+    save_semantic_cache([{"id": "n", "source_file": "doc.md"}], [], root=tmp_path)
+
+    # Corrupt the on-disk entry (e.g. an old producer wrote unescaped
+    # backslashes, or a partial write left truncated JSON).
+    h = file_hash(f, tmp_path)
+    entry = cache_dir(tmp_path, "semantic") / f"{h}.json"
+    assert entry.exists()
+    entry.write_text('{"nodes": [ this is not valid json')
+
+    with pytest.warns(RuntimeWarning, match="corrupt"):
+        nodes, _, _, uncached = check_semantic_cache([str(f)], root=tmp_path)
+
+    # The corrupt entry is a miss, so the file is re-dispatched for extraction.
+    assert nodes == []
+    assert uncached == [str(f)]