README.md

---
dataset_info:
  features:
  - name: level
    dtype: string
  - name: user
    dtype: string
  - name: datetime
    dtype: timestamp[ns, tz=Europe/Berlin]
  - name: kind
    dtype: string
  - name: password
    dtype: string
  - name: success
    dtype: bool
  - name: guess
    dtype: string
  - name: prompt
    dtype: string
  - name: answer
    dtype: string
  - name: llm
    dtype: string
  - name: raw_answer
    dtype: string
  - name: defender_time_sec
    dtype: float64
  - name: blocked_by
    dtype: string
  - name: level_order
    sequence: int64
  - name: setup
    dtype: string
  - name: defense
    dtype: string
  - name: embedding
    sequence: float32
  - name: prediction_direct
    dtype: float64
  - name: binary_prediction_direct
    dtype: bool
  - name: prediction_partial
    dtype: float64
  - name: binary_prediction_partial
    dtype: bool
  - name: prediction_indirect
    dtype: float64
  - name: binary_prediction_indirect
    dtype: bool
  - name: prediction_input_obfuscation
    dtype: float64
  - name: binary_prediction_input_obfuscation
    dtype: bool
  - name: prediction_output_obfuscation
    dtype: float64
  - name: binary_prediction_output_obfuscation
    dtype: bool
  - name: prediction_output_fiction
    dtype: float64
  - name: binary_prediction_output_fiction
    dtype: bool
  - name: prediction_input_non_english
    dtype: float64
  - name: binary_prediction_input_non_english
    dtype: bool
  - name: prediction_persuasion
    dtype: float64
  - name: binary_prediction_persuasion
    dtype: bool
  - name: prediction_context_override
    dtype: float64
  - name: binary_prediction_context_override
    dtype: bool
  - name: salient_category
    dtype: string
  splits:
  - name: train
    num_bytes: 247031328
    num_examples: 36194
  download_size: 254820537
  dataset_size: 247031328
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---