Upload Code_files/uslap.py with huggingface_hub

Code_files/uslap.py

@@ -158,8 +158,11 @@ def query(text):
     """Run a single query through the AMR pipeline. Returns formatted string.
 
     PRIORITY:
-    1. Search DB. If found, return DB data.
-    2. Compute via amr_dhakaa (handles ALL intents including tasrif).
+    1. Search DB (term_nodes index). If found, return DB data only.
+       Do NOT append hypothesis output to a DB hit.
+    2. If input is transliterated AA (Latin chars), attempt root lookup
+       via transliteration map before falling to hypothesis mode.
+    3. Not found → return NOT_FOUND string (no hypothesis generation).
 
     Fixed 2026-03-29: was going straight to think() and skipping
     handler.search(), which meant indexed entries (scholars, peoples,
@@ -167,6 +170,9 @@ def query(text):
 
     Fixed 2026-04-02: tasrif now routes through dhakaa pipeline
     (perceive → reason → articulate) instead of bypass.
+
+    Fixed 2026-04-15: DB hits no longer trigger hypothesis append.
+    Transliterated AA queries route to root lookup, not hypothesis.
     """
     from uslap_handler import search
 
@@ -174,12 +180,8 @@ def query(text):
     search_result = search(text)
 
     if search_result['found'] and search_result['nodes']:
-        # Found in DB — format and return the DB data
-        # But ALSO run through think() for root computation
-        # The DB data takes priority in the output
+        # Found in DB — return DB data ONLY. No hypothesis appended.
         lines = []
-
-        # Group by source table
         by_table = {}
         for node in search_result['nodes']:
             tbl = node['source_table']
@@ -193,31 +195,30 @@ def query(text):
 
         for tbl, nodes in by_table.items():
             lines.append(f"║ [{tbl}]:")
-            for n in nodes[:10]:  # Limit per table
+            for n in nodes[:10]:
                 term = n['term'] if isinstance(n, dict) else n[1]
                 score = n['score'] if isinstance(n, dict) else n[7]
                 src_id = n['source_id'] if isinstance(n, dict) else n[5]
                 lines.append(f"║   {term} (id={src_id}, score={score})")
 
         lines.append("╚" + "═" * 58 + "╝")
+        return '\n'.join(lines)
 
-        db_output = '\n'.join(lines)
-
-        # Step 2: Also run AMR computation for root analysis
-        from amr_dhakaa import think
-        result = think(text)
-        amr_output = result['output']
-
-        # Don't append "No entries found" if DB already returned results
-        if 'No entries found' in amr_output:
-            return db_output
-        else:
-            return db_output + '\n\n' + amr_output
-
-    # Not found in DB — fall through to AMR computation only
+    # Step 2: transliterated AA? attempt root lookup via dhakaa perceive layer
+    # before allowing hypothesis generation. dhakaa handles intent routing.
     from amr_dhakaa import think
     result = think(text)
-    return result['output']
+    output = result.get('output', '')
+
+    # Step 3: if dhakaa produced only hypothesis output, suppress it.
+    # "Not found in DB" is the correct answer — hypothesis is unconfirmed.
+    # The hypothesis block can start on line 1 or line 2 (after a ╔══ border).
+    output_lines = output.splitlines()
+    for line in output_lines[:4]:
+        if 'HYPOTHESIS:' in line:
+            return f"Not found in DB: '{text}'"
+
+    return output
 
 
 def full_report(text):