| | --- |
| | language: |
| | - en |
| | task_categories: |
| | - reinforcement-learning |
| | - text-generation |
| | tags: |
| | - eval |
| | - safety |
| | - coherence |
| | - alignment |
| | - recovery |
| | license: other |
| | size_categories: |
| | - small |
| | pretty_name: Clarus Alignment Recovery Dynamics v0.1 |
| | --- |
| | |
| | Clarus Alignment Recovery Dynamics v0.1 |
| |
|
| | This dataset measures recovery after an alignment flip. |
| |
|
| | Focus |
| |
|
| | - Not only whether a system flips |
| | - But whether it can recover |
| | - And whether it relapses under renewed pressure |
| |
|
| | Design |
| |
|
| | - One row per step |
| | - Steps form a trajectory grouped by case_id |
| | - A recovery window defines how quickly recovery must occur |
| | |
| | Columns |
| | |
| | - flip_signal_expected |
| | none, early_warning, flip, cascade |
| |
|
| | - first_flip_step_expected |
| | First step where a flip is expected, or -1 |
| | |
| | - recovery_expected |
| | true if a recovery probe exists |
| |
|
| | - recovery_quality_expected |
| | none, weak, partial, full |
| |
|
| | - recovery_window_steps |
| | Number of steps allowed to reach recovery quality |
| |
|
| | Scoring idea |
| |
|
| | - pass_recovered_full |
| | - pass_recovered_partial |
| | - fail_no_recovery |
| | - fail_relapse |
| | |
| | Intended use |
| | |
| | - Evaluate model resilience under authority pressure |
| | - Compare recovery speed and stability across models |
| | - Detect relapse behavior under renewed incentive pressure |
| | |
