annotation_task_name: Agent Trace Comparison
data_files:
- data/agent-comparisons.json
item_properties:
  id_key: id
  text_key: task_description
task_dir: .
output_annotation_dir: annotation_output
output_annotation_format: json
instance_display:
  layout:
    direction: horizontal
    gap: 16px
  fields:
  - key: trace_a
    type: dialogue
    label: Agent A
    display_options:
      show_turn_numbers: true
      alternating_shading: true
  - key: trace_b
    type: dialogue
    label: Agent B
    display_options:
      show_turn_numbers: true
      alternating_shading: true
annotation_schemes:
- annotation_type: radio
  name: overall_preference
  description: Which agent performed better overall?
  labels:
  - name: agent_a_much_better
    tooltip: Agent A was clearly better
  - name: agent_a_slightly_better
    tooltip: Agent A was slightly better
  - name: tie
    tooltip: Both were equally good
  - name: agent_b_slightly_better
    tooltip: Agent B was slightly better
  - name: agent_b_much_better
    tooltip: Agent B was clearly better
  sequential_key_binding: true
- annotation_type: radio
  name: completeness
  description: Which agent provided a more complete answer?
  labels:
  - name: agent_a
  - name: tie
  - name: agent_b
  sequential_key_binding: true
- annotation_type: radio
  name: efficiency
  description: Which agent was more efficient (fewer unnecessary steps)?
  labels:
  - name: agent_a
  - name: tie
  - name: agent_b
  sequential_key_binding: true
- annotation_type: radio
  name: accuracy
  description: Which agent's output was more accurate?
  labels:
  - name: agent_a
  - name: tie
  - name: agent_b
  sequential_key_binding: true
- annotation_type: radio
  name: helpfulness
  description: Which agent's final response was more helpful to the user?
  labels:
  - name: agent_a
  - name: tie
  - name: agent_b
  sequential_key_binding: true
- annotation_type: multirate
  name: dimension_ratings
  description: Rate each agent on these dimensions (1=Poor, 5=Excellent)
  options:
  - name: agent_a_correctness
    label: Agent A - Correctness
  - name: agent_a_efficiency
    label: Agent A - Efficiency
  - name: agent_a_communication
    label: Agent A - Communication
  - name: agent_b_correctness
    label: Agent B - Correctness
  - name: agent_b_efficiency
    label: Agent B - Efficiency
  - name: agent_b_communication
    label: Agent B - Communication
  labels:
  - '1'
  - '2'
  - '3'
  - '4'
  - '5'
- annotation_type: text
  name: preference_reason
  description: Why do you prefer one agent over the other?
user_config:
  allow_all_users: true
require_no_password: true