phase 3 day 5: expand wikitext to 20 questions → RLV 19/20

Phase A robustness verification: doubled the test set from 10 to 20.

New questions (Q11-Q20):
  Q11: Boulter's Casualty character → PASS (Kieron Fletcher)
  Q12: How to Curse theatre → PASS (Bush Theatre)
  Q13: Boulter's character in Blackburn's film → PASS (Sean) [multi-hop]
  Q14: Du Fu birth year → PASS (712)
  Q15: Du Fu's children count → PASS (five)
  Q16: Du Fu's Sichuan destination → PASS (Chengdu)
  Q17: Haiku poet influenced by Du Fu → PASS (Matsuo Basho)
  Q18: Du Fu "poet historian" epithet → FAIL [multi-hop]
  Q19: Kiss You Billboard Hot 100 peak → PASS (46)
  Q20: The Independent on Simon Cowell → PASS [multi-hop]

Results:
  Original 10: 10/10 (maintained)
  New 10:       9/10 (Q18 failed — "poet historian" epithet not found)
  Total:       19/20 (95%)

The single failure (Q18) is a difficult multi-hop question requiring
the model to understand that a Chinese literary criticism term maps
to the English translation "poet historian". This is a translation/
cultural knowledge gap, not a retrieval failure.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: unamedkr

bench/rlv/eval/eval_wikitext.py

@@ -106,6 +106,64 @@
         "question": "Who directed the Kiss You music video?",
         "fragments": ["vaughan arnell", "arnell"],
     },
+
+    # === Phase A expanded set: 10 more questions (Q11-Q20) ===
+
+    # Boulter — deeper facts
+    {
+        "id": 11, "topic": "boulter", "type": "single-hop",
+        "question": "What character did Boulter play in the TV series Casualty?",
+        "fragments": ["kieron fletcher", "kieron", "fletcher"],
+    },
+    {
+        "id": 12, "topic": "boulter", "type": "single-hop",
+        "question": "At which theatre was the play How to Curse performed?",
+        "fragments": ["bush theatre", "bush"],
+    },
+    {
+        "id": 13, "topic": "boulter", "type": "multi-hop",
+        "question": "What character did Boulter play in the film directed by Olly Blackburn?",
+        "fragments": ["sean"],
+    },
+
+    # Du Fu — deeper facts
+    {
+        "id": 14, "topic": "dufu", "type": "single-hop",
+        "question": "In what year was Du Fu born?",
+        "fragments": ["712"],
+    },
+    {
+        "id": 15, "topic": "dufu", "type": "single-hop",
+        "question": "How many children did Du Fu have by 757?",
+        "fragments": ["five", "5"],
+    },
+    {
+        "id": 16, "topic": "dufu", "type": "single-hop",
+        "question": "Which city did Du Fu move to in December, fleeing to Sichuan province?",
+        "fragments": ["chengdu"],
+    },
+    {
+        "id": 17, "topic": "dufu", "type": "single-hop",
+        "question": "Which famous haiku poet was strongly influenced by Du Fu?",
+        "fragments": ["matsuo basho", "basho", "bash"],
+    },
+    {
+        "id": 18, "topic": "dufu", "type": "multi-hop",
+        "question": "What two-word epithet meaning 'poet historian' did Chinese critics give Du Fu?",
+        "fragments": ["poet historian", "historian"],
+    },
+
+    # Kiss You — deeper facts
+    {
+        "id": 19, "topic": "kiss_you", "type": "single-hop",
+        "question": "At what number did Kiss You peak on the US Billboard Hot 100?",
+        "fragments": ["46"],
+    },
+    {
+        "id": 20, "topic": "kiss_you", "type": "multi-hop",
+        "question": "According to The Independent, who reported that Simon Cowell wanted Kiss You as the lead single?",
+        "fragments": ["independent"],
+    },
 ]