top keys ['scores', 'cases'] scores {'Storyline Consistency': 66.16666666666667, 'Anthropomorphism': 52.833333333333336, 'Character Fidelity': 51.166666666666664, 'Storyline Quality': 79.5, 'avg': 62.416666666666664, 'bleu': 0.02527484677136496, 'rouge_l': 0.08387797463409082}
cases 3 sim entries 3 test_set entries 200
case keys ['simulation', 'simulation_str', 'score', 'critique'] critique keys ['Storyline Consistency', 'Anthropomorphism', 'Character Fidelity', 'Storyline Quality', 'bleu', 'rouge_l']
dim keys ['flaws', 'score']

dimension scores recomputed 12
  paper main text: 100-5S+1.5R clip[0,100]: 12/12
  official exact: 100-(S-0.3R)*5 clip[0,100]: 12/12
  same, no upper clip (paper-era): 12/12
  same, no clip at all: 12/12
  appendix: 100-S+1.5R: 0/12
  no length correction: 100-5S: 0/12
  R counts env too: 100-5S+1.5A clip: 0/12
  lambda=1.0: 0/12
  lambda=2.0: 0/12
  factor 4: 0/12
  factor 6: 0/12
severity value histogram [(2, 12), (3, 24), (4, 6), (5, 1)]
severity types Counter({'int': 43})
upper clip hit {} lower clip hit {}
R per case: min/mean/max 5 9.666666666666666 13 histogram [(5, 1), (11, 1), (13, 1)]
non-NSP messages per case: min/mean/max 8 12.666666666666666 16
 mean Storyline Consistency 66.16666666666667 file 66.16666666666667
 mean Anthropomorphism 52.833333333333336 file 52.833333333333336
 mean Character Fidelity 51.166666666666664 file 51.166666666666664
 mean Storyline Quality 79.5 file 79.5
 avg of dim means 62.41666666666667 file avg 62.416666666666664
 mean of case scores 62.416666666666664
 case score == mean of its four: 3 / 3
 mean bleu 0.02527484677136496 file 0.02527484677136496
 mean rouge_l 0.08387797463409082 file 0.08387797463409082

linear fit vanilla (×5, no correction) on R, this sample only: slope -0.7452 intercept 55.1202 n 3
mean of 100-S (no ×5) on this sample 89.58  mean of 100-5S 47.92 ; paper regression intercept 59.06 slope -1.59

simulation keys ['book_title', 'i_p', 'i_c', 'circumstance', 'simulation', 'involved_character_profiles']
role counts Counter({'NSP': 38, 'char': 29, 'Env': 9}) speaker messages with own name prefix 38 / 38
case keys equal between sim and eval: True
how many eval cases are in test_set.json 0 of 3
first indices of those cases in test_set []
key_characters fields in the simulation file circumstance ['motivation', 'name']