table,source,model,dataset,threshold_selection,examples,positives,negatives,threshold,recall,false_positive_rate,roc_auc,average_precision,true_positive,false_positive
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R3 reference,civil_reference.toxicity,own development subset,750,150,600,0.7053980718999331,0.5133333333333333,0.05,0.8729,0.6623435228143205,77,30
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R3 reference,wiki_native.toxicity,own development subset,663,369,294,0.4509108293116966,0.943089430894309,0.034013605442176874,0.9858414910679719,0.9905171829010868,348,10
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R3 reference,wiki_native.harassment,own development subset,401,200,201,0.47529342292329135,0.95,0.04975124378109453,0.9859452736318408,0.9844993359176231,190,10
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R7 lr-1e-05 final,civil_reference.toxicity,own development subset,750,150,600,0.4702263687414845,0.5133333333333333,0.05,0.8738111111111111,0.6874052765853308,77,30
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R7 lr-1e-05 final,wiki_native.toxicity,own development subset,663,369,294,0.2778274756540858,0.9349593495934959,0.04421768707482993,0.9837306196191213,0.9879333741687993,345,13
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R7 lr-1e-05 final,wiki_native.harassment,own development subset,401,200,201,0.6541404835998934,0.905,0.04975124378109453,0.9810696517412936,0.9787636299441846,181,10
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R7 lr-3e-05 final,civil_reference.toxicity,own development subset,750,150,600,0.48639491099281995,0.49333333333333335,0.05,0.8776555555555555,0.6806104337788443,74,30
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R7 lr-3e-05 final,wiki_native.toxicity,own development subset,663,369,294,0.28005849203697963,0.926829268292683,0.047619047619047616,0.9822834282764596,0.9858958287240037,342,14
R3 reference and R7 final checkpoints,risk-recovery-20260924/r7/summary.json,R7 lr-3e-05 final,wiki_native.harassment,own development subset,401,200,201,0.747817102949135,0.925,0.04477611940298507,0.9780099502487563,0.9696533483808822,185,9
Fixed lexical control on the same R7 toxicity examples,risk-recovery-20260924/full-replay-lexical-control/summary.json,same-data lexical control,civil_reference,Civil development threshold (unchanged),750,150,600,0.25537019968032837,0.8333333333333334,0.04833333333333333,0.9545333333333333,0.8936549256734655,125,29
Fixed lexical control on the same R7 toxicity examples,risk-recovery-20260924/full-replay-lexical-control/summary.json,same-data lexical control,wiki_at_civil_threshold,Civil development threshold (unchanged),663,369,294,0.25537019968032837,0.9349593495934959,0.034013605442176874,0.9921095809597552,0.9943349900402629,345,10
Fixed lexical control on the same R7 toxicity examples,risk-recovery-20260924/full-replay-lexical-control/summary.json,same-data lexical control,wiki_native,own Wiki development subset,663,369,294,0.21526019275188446,0.9566395663956639,0.03741496598639456,0.9921095809597552,0.9943349900402629,353,11
Fixed lexical control on the same R7 toxicity examples,risk-recovery-20260924/full-replay-lexical-control/summary.json,same-data lexical control,context:all,Civil development threshold (unchanged),1373,526,847,0.25537019968032837,0.9467680608365019,0.18181818181818182,0.9545903457068338,0.9286604504200875,498,154
Fixed lexical control on the same R7 toxicity examples,risk-recovery-20260924/full-replay-lexical-control/summary.json,same-data lexical control,context:source:civil_comments,Civil development threshold (unchanged),883,356,527,0.25537019968032837,0.9550561797752809,0.06451612903225806,0.9852621367503145,0.9827724519469723,340,34
Fixed lexical control on the same R7 toxicity examples,risk-recovery-20260924/full-replay-lexical-control/summary.json,same-data lexical control,context:source:sbic_v2,Civil development threshold (unchanged),479,170,309,0.25537019968032837,0.9294117647058824,0.3883495145631068,0.8670474014848658,0.7725693884621248,158,120
Fixed lexical control on the same R7 toxicity examples,risk-recovery-20260924/full-replay-lexical-control/summary.json,same-data lexical control,context:source:wiki_detox,Civil development threshold (unchanged),11,0,11,0.25537019968032837,,0.0,,,0,0