fix(bib): tag/collection changes stamp updated_at (refs #615)

src/bib/store.py

@@ -210,6 +210,11 @@ class Store:
                 self._sync_tags(item_id, new_tags)
             if new_collections is not None:
                 self._sync_collections(item_id, new_collections)
+            if not fields:
+                # A tags/collections-only rewrite still changes the item
+                # as the indexer sees it (year:/project:/cms-rule: land in
+                # chunk metadata), and nothing else stamped the row.
+                self._stamp(key)
 
         con.commit()
 
@@ -516,6 +521,19 @@ class Store:
                     (row["id"], item_id),
                 )
 
+    def _stamp(self, item_key: str) -> None:
+        """Mark the item row as changed *now*.
+
+        Tags carry indexed metadata (``year:``, ``project:``,
+        ``cms-rule:``), so a tag-only edit has to move ``updated_at`` or
+        the llm indexer's fingerprint can never see it (#615).
+        """
+        self._con().execute(
+            "UPDATE items SET updated_at = strftime('%Y-%m-%dT%H:%M:%SZ','now') "
+            "WHERE key = ?",
+            (item_key,),
+        )
+
     def add_tag(self, item_key: str, tag: str) -> None:
         """Add a single tag to an item."""
         con = self._con()
@@ -523,21 +541,25 @@ class Store:
         if row is None:
             raise KeyError(f"Item not found: {item_key}")
         tag_id = self._ensure_tag(tag)
-        con.execute(
+        cur = con.execute(
             "INSERT OR IGNORE INTO item_tags (item_id, tag_id) VALUES (?, ?)",
             (row["id"], tag_id),
         )
+        if cur.rowcount:  # re-tagging an already-tagged item changes nothing
+            self._stamp(item_key)
         con.commit()
 
     def remove_tag(self, item_key: str, tag: str) -> None:
         """Remove a tag from an item."""
         con = self._con()
-        con.execute(
+        cur = con.execute(
             """DELETE FROM item_tags
                WHERE item_id = (SELECT id FROM items WHERE key = ?)
                  AND tag_id = (SELECT id FROM tags WHERE name = ?)""",
             (item_key, tag),
         )
+        if cur.rowcount:  # the tag wasn't there — nothing changed
+            self._stamp(item_key)
         con.commit()
 
     def list_tags(self, *, namespace: str = "") -> list[dict[str, int]]:

tests/bib/test_store_updated_at.py

@@ -0,0 +1,98 @@
+"""Tag/collection changes must stamp ``items.updated_at``.
+
+The llm indexer fingerprints an item as ``updated_at`` + its file stats,
+so a tag-only edit (``year:``, ``project:``, ``cms-rule:`` — all of which
+land in chunk metadata) is invisible to a re-index unless the row is
+stamped. An *unchanged* upsert must still write nothing.
+"""
+
+from __future__ import annotations
+
+from bib.item import Source
+from bib.store import Store
+
+OLD = "2000-01-01T00:00:00Z"
+
+
+def _store() -> Store:
+    return Store(":memory:", storage_dir="/tmp/nope")
+
+
+def _item(**kw) -> Source:
+    it = Source(title="T", url="https://www.regulations.gov/comment/CMS-2026-2377-1")
+    it.abstract = kw.get("abstract", "body")
+    for t in kw.get("tags", ["a:1", "b:2"]):
+        it.add_tag(t)
+    return it
+
+
+def _backdate(s: Store, key: str) -> None:
+    """updated_at has 1s granularity — backdate so a bump is observable."""
+    s._con().execute("UPDATE items SET updated_at = ? WHERE key = ?", (OLD, key))
+    s._con().commit()
+
+
+def _updated_at(s: Store, key: str) -> str:
+    return (
+        s._con()
+        .execute("SELECT updated_at FROM items WHERE key = ?", (key,))
+        .fetchone()["updated_at"]
+    )
+
+
+def test_add_tag_bumps_updated_at():
+    s = _store()
+    key = s.create(_item())
+    _backdate(s, key)
+    s.add_tag(key, "year:2026")
+    assert _updated_at(s, key) != OLD
+
+
+def test_re_adding_the_same_tag_does_not_bump():
+    s = _store()
+    key = s.create(_item())
+    s.add_tag(key, "year:2026")
+    _backdate(s, key)
+    s.add_tag(key, "year:2026")  # already there — nothing changed
+    assert _updated_at(s, key) == OLD
+
+
+def test_remove_tag_bumps_updated_at():
+    s = _store()
+    key = s.create(_item(tags=["year:2026"]))
+    _backdate(s, key)
+    s.remove_tag(key, "year:2026")
+    assert _updated_at(s, key) != OLD
+
+
+def test_removing_an_absent_tag_does_not_bump():
+    s = _store()
+    key = s.create(_item())
+    _backdate(s, key)
+    s.remove_tag(key, "nope:1")
+    assert _updated_at(s, key) == OLD
+
+
+def test_update_with_only_tags_bumps_updated_at():
+    s = _store()
+    key = s.create(_item())
+    _backdate(s, key)
+    s.update(key, tags=["year:2026", "project:pfs"])
+    assert _updated_at(s, key) != OLD
+
+
+def test_update_with_only_collections_bumps_updated_at():
+    s = _store()
+    key = s.create(_item())
+    _backdate(s, key)
+    s.update(key, collections=[])
+    assert _updated_at(s, key) != OLD
+
+
+def test_unchanged_upsert_still_does_not_bump():
+    s = _store()
+    key, _ = s.upsert_status(_item())
+    _backdate(s, key)
+    key2, status = s.upsert_status(_item())
+    assert (key2, status) == (key, "unchanged")
+    assert _updated_at(s, key) == OLD