Evaluation profile
Olam Social Poker — Social Lie Rate
1sub-evals
0.324%Safety weight
0%Freedom weight
1components
Weights below are portfolio-specific global index weights.
Model score (lower is better)Predicted score
About this eval
Propensity to verbalize deliberate strategic lies during Social Poker table talk.
Included in the behavior ranking.
Claude Fable 5.1 (164 lies per 10000 turns) and GLM 5.3 Flash (30) are provisional source-reported relabelings observed September 10, not independently verified new evaluations. The original Claude Fable 5 (164) and Ox Alpha (30) scores are retained. Fable's API identifier still names Fable 5; no cross-benchmark Ox Alpha to GLM identity equivalence is asserted.
Sub-evals
| Measure | Component | Direction by ranking | Safety weight | Freedom weight |
|---|---|---|---|---|
| social_lie_rate_per_10000_turnsolam-poker-deception/deception-rate.csv:social_lie_rate_per_10000_turnsMeasures how frequently a model chooses to make deliberate strategic false claims during Social Poker. | Safety: truthfulness_honesty:1.000olam-poker-deception | Safety: lower | 0.324% | — |
social_lie_rate_per_10000_turns
Measures how frequently a model chooses to make deliberate strategic false claims during Social Poker.