Checkpoint frontend UI/UX overhaul and ingestion embed benchmark work

apps/ai-service/tests/test_bootstrap.py

@@ -0,0 +1,38 @@
+"""`_catalog_names` — the drug names shown to `LlmQueryUnderstander`.
+
+Reproduces a live 2026-08-06 bug: picking the first N aliases alphabetically
+could drop a drug's own recognizable name entirely, breaking multi-turn
+follow-ups where the drug is no longer restated in the raw turn text (see
+`docs/progress-log.md` for the exact failure: "Liều paracetamol cho trẻ em"
+-> two clarify rounds -> "Không tìm thấy paracetamol").
+"""
+from bootstrap import _catalog_names
+
+
+def test_alphabetically_early_junk_alias_does_not_bury_the_canonical_name():
+    aliases = {
+        "paracetamol_acetaminophen": {
+            "0Frezefev", "ABAB", "Ace kid 80", "PARACETAMOL", "Acetaminophen",
+        },
+    }
+    shown = _catalog_names(aliases)["paracetamol_acetaminophen"]
+    assert "paracetamol acetaminophen" in shown
+
+
+def test_canonical_name_is_always_first():
+    aliases = {"metformin": {"METFORMIN", "Axiol", "Dybis", "Zzyzx"}}
+    shown = _catalog_names(aliases)["metformin"]
+    assert shown.split(", ")[0] == "metformin"
+
+
+def test_drug_with_no_aliases_still_shows_its_canonical_name():
+    shown = _catalog_names({"some_drug": set()})["some_drug"]
+    assert shown == "some drug"
+
+
+def test_output_is_capped_and_does_not_duplicate_the_canonical_name():
+    aliases = {"drug_x": {f"Brand{i}" for i in range(20)} | {"DRUG X", "drug x"}}
+    shown = _catalog_names(aliases)["drug_x"].split(", ")
+    assert shown[0] == "drug x"
+    assert shown.count("drug x") == 1
+    assert len(shown) <= 3