fix(mp2): aligned tables and ASCII boxes are layout, not context stuffing

src/skillspector/nodes/analyzers/static_patterns_memory_poisoning.py

@@ -79,6 +79,15 @@
     ),
 ]
 
+# Layout filler: characters that repeat because something is being *aligned*, not because
+# context is being displaced — aligned table cells, ASCII boxes, horizontal rules, padded
+# columns built by UI code. Box drawing is U+2500-U+257F.
+_MP2_LAYOUT_FILLER = re.compile(r"^[\s\-=_*.|:+~\u2500-\u257F]+$")
+
+# Above this width the run stops being plausible formatting: nobody reads a table row or an
+# ASCII box 400 characters wide, so a repetition that long is reported even if it is filler.
+_MP2_LAYOUT_MAX_RUN = 400
+
 # MP2: Context Window Stuffing — filling context to displace content
 MP2_PATTERNS = [
     (r"(.{2,20}?)\1{20,}", 0.8),
@@ -186,6 +195,14 @@ def ctx(start: int) -> str:
             non_ws_chars = set(captured) - {" ", "\t", "\n", "\r"}
             if len(non_ws_chars) <= 1 and not any(c in captured for c in (" ", "\t")):
                 continue
+            # Same reasoning as the guard above, extended to the case it misses: a unit made
+            # only of whitespace and alignment characters repeats because a table, a box or a
+            # column is being drawn. Runs wider than a readable line are still reported.
+            if (
+                _MP2_LAYOUT_FILLER.match(captured)
+                and len(match.group(0)) < _MP2_LAYOUT_MAX_RUN
+            ):
+                continue
             line_num = get_line_number(content, match.start())
             findings.append(
                 AnalyzerFinding(

tests/unit/test_patterns_new.py

@@ -551,6 +551,34 @@ def test_mp2_repeated_pattern(self) -> None:
     def test_mp2_separator_not_flagged(self) -> None:
         assert not any(f.rule_id == "MP2" for f in mp_mod.analyze("=" * 80, "test.md", "markdown"))
 
+    def test_mp2_aligned_markdown_table_is_layout(self) -> None:
+        # An aligned table repeats spaces and dashes because cells are being lined up, not
+        # because context is being displaced.
+        content = (
+            "| campo      | quando            | significato                        |\n"
+            "| ---------- | ----------------- | ---------------------------------- |\n"
+            "| id         | sempre            | identificatore stabile del record  |\n"
+            "| aggiornato | dopo ogni scrittura | timestamp dell'ultima modifica   |\n"
+        )
+        assert not any(f.rule_id == "MP2" for f in mp_mod.analyze(content, "docs/tabella.md", "markdown"))
+
+    def test_mp2_ascii_box_is_layout(self) -> None:
+        content = "│  └───────────────────────────────────────────────────────┘  │\n"
+        assert not any(f.rule_id == "MP2" for f in mp_mod.analyze(content, "README.md", "markdown"))
+
+    def test_mp2_padded_column_in_code_is_layout(self) -> None:
+        content = 'lines.append(f"{Colors.DIM}|{Colors.RESET}                                   ")\n'
+        assert not any(f.rule_id == "MP2" for f in mp_mod.analyze(content, "src/ui.py", "python"))
+
+    def test_mp2_whitespace_run_beyond_layout_width_is_still_stuffing(self) -> None:
+        # Layout stops being a plausible explanation well before this width.
+        content = "testo" + " " * 800 + "altro testo\n"
+        assert any(f.rule_id == "MP2" for f in mp_mod.analyze(content, "test.md", "markdown"))
+
+    def test_mp2_repeated_content_is_still_stuffing(self) -> None:
+        # The unit repeated here is content, not alignment: unchanged behaviour.
+        assert any(f.rule_id == "MP2" for f in mp_mod.analyze("lorem " * 60, "test.md", "markdown"))
+
     @pytest.mark.parametrize(
         "content",
         [