Update app.py

app.py

@@ -1492,9 +1492,7 @@ async def browse_subject(req: SubjectRequest, request: Request):
         }}
 
     # Hard-filter against the real vocabulary. A name the model invented is
-    # discarded here rather than being handed to the search, where it would
-    # fail the category test and fall through to a broad area match that looks
-    # like a real answer.
+    # discarded here rather than handed to the search.
     bad_category = bool(category) and not is_real_category(category)
     bad_area     = bool(area)     and not is_real_area(area)
     if bad_category:
@@ -1506,27 +1504,54 @@ async def browse_subject(req: SubjectRequest, request: Request):
                        area, req.subject)
         area = ""
 
-    if not category and not area:
-        usage.usage.record_gap("journal_subject", "subject_unrecognised")
+    # Suggestions the user can act on: the model's nearest names, hard-filtered
+    # to real ones, topped up deterministically. Never invented.
+    def _suggestions() -> list:
+        out, seen = [], set()
+        for n in (parsed.get("nearest_categories") or []):
+            n = str(n).strip().lower()
+            if n and n not in seen and is_real_category(n):
+                seen.add(n); out.append(n)
+        for n in suggest_categories(req.subject.strip(), limit=6):
+            if n not in seen:
+                seen.add(n); out.append(n)
+        counts = subject_vocabulary()["categories"]
+        return [{"category": n.title(), "n_journals": counts.get(n, 0)} for n in out[:6]]
+
+    # An AREA is only a legitimate answer when the user asked for one. If the
+    # model named a category and that category was not real, the user asked for
+    # something specific that this taxonomy does not have — and answering with
+    # 30 top journals from the surrounding area is a confidently wrong answer,
+    # not a partial one. "Crime" is not served by showing Tourism Management.
+    #
+    # Refusing and naming the nearest real categories is both the honest answer
+    # and the more useful one: Law, Sociology and Political Science, and
+    # Pathology and Forensic Medicine are where criminology research is indexed.
+    if bad_category or (not category and not area):
+        usage.usage.record_gap(
+            "journal_subject",
+            "category_invented" if bad_category else "subject_unrecognised")
         request.state.usage_recorded = True
-        near = suggest_categories(req.subject.strip(), limit=6)
-        hint = ("  Closest SCImago categories: "
-                + ", ".join(n.title() for n in near) + ".") if near else ""
+        sugg = _suggestions()
+        base = (f"SCImago and Scopus have no subject category for "
+                f"'{req.subject.strip()}'.")
+        hint = ("  The closest indexed categories are: "
+                + ", ".join(s["category"] for s in sugg) + ".") if sugg else ""
         return {"result": {
-            "confidence": "not_found", "normalised": normalised,
-            "user_message": (user_message or
-                f"SCImago has no subject category matching '{req.subject.strip()}'."
-                + hint),
-            "journals": [],
+            "confidence":  "not_found",
+            "normalised":  normalised,
+            "user_message": (user_message + "  " if user_message else "") + base + hint,
+            "suggestions": sugg,
+            "journals":    [],
         }}
 
     journals, match_level = search_by_subject(category, area, top_n=30, with_level=True)
 
-    # match_level 1 means nothing matched the category and these journals were
-    # selected by AREA alone — thousands of them, cut to 30 by h-index. That is
-    # a browsing aid, not an answer, and it must be labelled as one. The old
-    # `fallback_used` flag never fired for this case because the first search
-    # already returned a full page.
+    # Reaching here with match_level 1 now means only one thing: the model
+    # named NO category and a real area, i.e. the user asked for the area.
+    # "Show me top Engineering journals" is a legitimate request and this is
+    # the correct answer to it — still labelled, because 30 rows out of
+    # thousands ranked by h-index is a browsing aid, not a recommendation.
     fallback_used = (match_level == 1)
 
     if not journals:
@@ -1561,9 +1586,10 @@ async def browse_subject(req: SubjectRequest, request: Request):
 
     msg = user_message
     if fallback_used:
-        warn = (f"No SCImago category matched '{req.subject.strip()}'. "
-                f"These are the highest-ranked journals across the whole "
-                f"'{area.title()}' area, not a subject-specific list.")
+        warn = (f"'{req.subject.strip()}' is a subject AREA, not a specific "
+                f"category. These are the highest-ranked journals across the "
+                f"whole of '{area.title()}' by h-index — narrow to a category "
+                f"for a subject-specific list.")
         msg = f"{msg}  {warn}" if msg else warn
         usage.usage.record_gap("journal_subject", "area_fallback")
         request.state.usage_recorded = True