v6 A/B VERDICT scoring: an item passes only if both graders pass it

model              graders         nokit   kit60   kit61  rule1 (kit61 >= kit60-2)  kit61 vs nokit
Sonnet 5.5         gpt+gemini     31/39   35/39   34/39   ok   (-1)              +3
GPT-6-Luna         claude+gemini  28/39   35/39   37/39   ok   (+2)              +9
GPT-6-Sol          claude+gemini  30/39   34/39   34/39   ok   (+0)              +4
Gemini 3.8 Flash   claude+gpt     22/39   33/39   34/39   ok   (+1)              +12
Muse Spark 1.3     claude+gpt     27/39   27/39   29/39   ok   (+2)              +2

Rule 1 (every complete model kit61 >= kit60 - 2): HOLDS
Rule 2 (conform [eng] passed, 5 models; max 40): nokit 26  kit60 25  kit61 30; new kit61 build/test failures: none -> HOLDS
Rule 3 ('fine' items called blocking; max 20): nokit 1  kit60 0  kit61 0 -> HOLDS
Amendment 3 stable criterion (no complete model's kit61 below its nokit): HOLDS
Incomplete models: none
OUTCOME: v6.1.0 STABLE (Latest)

Per task (both graders; items passed nokit / kit60 / kit61):
  Sonnet 5.5         T1 4/4/4 of 6   T2 4/6/5 of 6   T3 6/7/7 of 7   T4 6/8/8 of 8   T5 5/5/5 of 6   T6 6/5/5 of 6
  GPT-6-Luna         T1 4/4/5 of 6   T2 4/5/6 of 6   T3 5/7/7 of 7   T4 7/8/8 of 8   T5 5/5/5 of 6   T6 3/6/6 of 6
  GPT-6-Sol          T1 4/4/5 of 6   T2 5/5/4 of 6   T3 6/7/7 of 7   T4 7/7/7 of 8   T5 5/6/6 of 6   T6 3/5/5 of 6
  Gemini 3.8 Flash   T1 5/4/5 of 6   T2 4/4/5 of 6   T3 5/7/6 of 7   T4 7/7/7 of 8   T5 0/5/5 of 6   T6 1/6/6 of 6
  Muse Spark 1.3     T1 6/4/3 of 6   T2 4/5/6 of 6   T3 4/5/7 of 7   T4 6/8/7 of 8   T5 5/5/5 of 6   T6 2/0/1 of 6

[eng] vs [kit] items (both graders; nokit / kit60 / kit61):
  Sonnet 5.5         eng 22/24/23     kit 9/11/11
  GPT-6-Luna         eng 22/24/27     kit 6/11/10
  GPT-6-Sol          eng 24/23/24     kit 6/11/10
  Gemini 3.8 Flash   eng 15/22/23     kit 7/11/11
  Muse Spark 1.3     eng 19/19/20     kit 8/8/9

Per grader, all items (nokit / kit60 / kit61), and grader agreement:
  Sonnet 5.5         gpt 36/36/36   gemini 31/36/35   agree 107/117
  GPT-6-Luna         claude 29/36/37   gemini 28/35/37   agree 115/117
  GPT-6-Sol          claude 33/37/38   gemini 30/34/34   agree 107/117
  Gemini 3.8 Flash   claude 28/37/38   gpt 22/33/34   agree 103/117
  Muse Spark 1.3     claude 31/29/34   gpt 27/28/29   agree 105/117

Rule 4 (reported, not scored): kit tokens per new-feature run (upper bound; 20k design budget)
  Sonnet 5.5         T5-kit60 19373   T5-kit61 17377   T6-kit60 6441   T6-kit61 6687
  GPT-6-Luna         T5-kit60 27604   T5-kit61 24228   T6-kit60 12218   T6-kit61 25777
  GPT-6-Sol          T5-kit60 22226   T5-kit61 29676   T6-kit60 25163   T6-kit61 21216
  Gemini 3.8 Flash   T5-kit60 19240   T5-kit61 17370   T6-kit60 9275   T6-kit61 16918
  Muse Spark 1.3     T5-kit60 14269   T5-kit61 24197   T6-kit60 6441   T6-kit61 13671