diff --git a/README.md b/README.md
index 9a7f111..7c72db9 100644
--- a/README.md
+++ b/README.md
@@ -58,17 +58,17 @@ That single command:
    one-shot indexing is never a problem.
 5. Validates the graph scope: no `test_*` addon modules indexed, no missing and no extra module
    roots compared to the expected module list, and **every indexed file carrying one of the
-   configured extensions** (`SOURCE_GLOBS`: `.py .xml .js .rst .md .css .scss .csv`) — anything
+   configured extensions** (`SOURCE_GLOBS`: `.py .xml .js .scss .csv .sql`) — anything
    else means `.cbmignore` did not apply. Counts per extension are logged, unexpected ones marked
    `(!)`:
 
    ```text
-   extensions_configured=.css,.csv,.js,.md,.py,.rst,.scss,.xml
+   extensions_configured=.csv,.js,.py,.scss,.sql,.xml
    extensions_indexed=5 files=12057
    extension .py      files=4949
    extension .xml     files=3276
    extension .json    files=39 (!)
-   unexpected_extensions=1 files=39 (not in SOURCE_GLOBS: .json)
+   unexpected_extensions=1 files=39 (not in the configured extensions: .json)
    ```
 
    Both checks read the graph with `query_graph` over its `File` nodes, not with `search_code`:
diff --git a/src/codebase_memory_deployv/indexer.py b/src/codebase_memory_deployv/indexer.py
index 3449564..9f25a52 100644
--- a/src/codebase_memory_deployv/indexer.py
+++ b/src/codebase_memory_deployv/indexer.py
@@ -39,7 +39,28 @@
 DEFAULT_PROJECT = "odoo_instance"
 CORE_DIRS = ("odoo/odoo",)
 CORE_PREFIX = "odoo/odoo/"
-SOURCE_GLOBS = ("*.py", "*.xml", "*.js", "*.rst", "*.md", "*.css", "*.scss", "*.csv", "*.sql")
+# Prose formats are deliberately absent. Measured on a real Odoo instance (188 modules,
+# 77935 DEFINES edges), every extension earns its place by how much of the graph it builds:
+#
+#     .py    2755 files  55603 definitions  20.2 per file  71.3% of the graph
+#     .xml   1883 files  14844 definitions   7.9 per file  19.0%
+#     .js     890 files   6368 definitions   7.2 per file   8.2%
+#     .scss   232 files    912 definitions   3.9 per file   1.2%
+#     .csv     86 files     86 definitions   1.0 per file   0.1%
+#     .sql     20 files     35 definitions   1.8 per file   0.0%
+#     .rst     73 files     73 definitions   1.0 per file   0.1%   <- dropped
+#     .css      8 files      8 definitions   1.0 per file   0.0%   <- dropped
+#     .md       2 files      6 definitions   3.0 per file   0.0%   <- dropped
+#
+# ".rst" and ".css" build exactly one node each — the File node itself, nothing inside it —
+# and ".md" only adds the headings of a README (every Section node in that graph came from
+# two markdown files). They are module prose, so the graph gains a path and no structure.
+#
+# ".csv" scores the same 1.0 and stays anyway: an indexed file is also greppable through
+# search_code, and ir.model.access.csv is a file people really do search. Extensions are a
+# weak lever on memory in any case (these three are 1.4% of the files, where .js is 15%),
+# so this is about keeping the graph free of prose, not about making a scope fit.
+SOURCE_GLOBS = ("*.py", "*.xml", "*.js", "*.scss", "*.csv", "*.sql")
 # What .cbmignore lets through, so what the graph is expected to hold: ".py", ".xml", ...
 SOURCE_EXTENSIONS = tuple(sorted(pattern[1:] for pattern in SOURCE_GLOBS))
 MANIFEST_NAMES = ("__manifest__.py", "__openerp__.py")
diff --git a/tests/test_indexer.py b/tests/test_indexer.py
index 82f88dc..5dc72e9 100644
--- a/tests/test_indexer.py
+++ b/tests/test_indexer.py
@@ -480,3 +480,26 @@ def test_check_layout_requires_odoo_bin(tmp_path):
     odoo_dir.mkdir()
     (odoo_dir / "odoo-bin").write_text("")
     assert indexer.check_layout(str(tmp_path)) == os.path.join(str(tmp_path), "odoo", "odoo-bin")
+
+
+def test_source_globs_exclude_module_prose():
+    """.rst/.md/.css build a File node and nothing inside it; the graph gains no structure."""
+    assert ".rst" not in indexer.SOURCE_EXTENSIONS
+    assert ".md" not in indexer.SOURCE_EXTENSIONS
+    assert ".css" not in indexer.SOURCE_EXTENSIONS
+
+
+def test_source_globs_keep_csv_for_text_search():
+    """ir.model.access.csv builds no structure either, but an indexed file stays greppable."""
+    assert ".csv" in indexer.SOURCE_EXTENSIONS
+
+
+def test_render_cbmignore_stops_letting_prose_through():
+    content = indexer.render_cbmignore(["extra_addons/vauxoo/sale_extended"])
+    assert "!extra_addons/vauxoo/sale_extended/**/*.py" in content
+    assert "!extra_addons/vauxoo/sale_extended/**/*.csv" in content
+    assert "*.rst" not in content
+    assert "*.md" not in content
+    assert "*.css" not in content
+    # ".scss" must survive a naive "*.css" removal: it ends with the same three letters.
+    assert "!extra_addons/vauxoo/sale_extended/**/*.scss" in content