metadata
dataset_info:
features:
- name: episode_id
dtype: string
- name: scenario_id
dtype: string
- name: messages
list:
- name: role
dtype: string
- name: content
dtype: string
- name: reward
dtype: float64
- name: reward_breakdown
struct:
- name: schema_version
dtype: string
- name: total
dtype: float64
- name: format_gate
dtype: bool
- name: components
struct:
- name: format_gate
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: valid
dtype: float64
- name: diagnosis
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: matrix_lookup
dtype: float64
- name: predicted
dtype: float64
- name: action_quality
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: no_secondary
dtype: float64
- name: cost_efficiency
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: total_cost
dtype: float64
- name: ratio
dtype: float64
- name: investigation
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: coverage
dtype: float64
- name: ordering
dtype: float64
- name: redundancy_penalty
dtype: float64
- name: min_ev_multiplier
dtype: float64
- name: minimal_evidence
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: min_set_used
dtype: float64
- name: extras
dtype: float64
- name: time
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: steps
dtype: float64
- name: excess
dtype: float64
- name: anti_gaming
struct:
- name: raw
dtype: float64
- name: weighted
dtype: float64
- name: weight
dtype: float64
- name: sub_scores
struct:
- name: no_info_penalty
dtype: float64
- name: quarantine_rate
dtype: float64
- name: quarantine_penalty
dtype: float64
- name: brier_bonus
dtype: float64
- name: counterfactual
dtype: 'null'
- name: format_gate_passed
dtype: bool
splits:
- name: train
num_bytes: 5257828
num_examples: 718
download_size: 5258953
dataset_size: 5257828
configs:
- config_name: default
data_files:
- split: train
path: data/train-*