**Name**: A string identifying the specific quality aspect being assessed

Examples: `"correctness"`, `"relevance_to_query"`, `"is_safe"`, `"guideline_adherence_politeness"`

**Value**: The actual result of the assessment, which can be:
- Numeric scores (e.g., `0.0` to `1.0`, `1` to `5`)
- Boolean values (`True`/`False`)
- Categorical labels (e.g., `"PASS"`, `"FAIL"`, `"EXCELLENT"`)
- Structured data (e.g., `{"score": 0.8, "confidence": 0.9}`)

**Rationale**: A string explaining why the assessment resulted in the given value

This explanation is crucial for transparency, debugging, and understanding evaluation behavior, especially for LLM-based assessments.

**LLM-based Evaluations**: Automated assessments using language models as judges
- Fast and scalable
- Can evaluate complex, subjective criteria
- Provide detailed reasoning in rationale

**Programmatic Checks**: Rule-based or algorithmic evaluations
- Deterministic and consistent
- Fast execution
- Good for objective, measurable criteria

**Human Reviews**: Manual assessments from human evaluators
- Highest quality for subjective evaluations
- Slower and more expensive
- Essential for establishing ground truth

All feedback types are treated equally in MLflow and can be combined to provide comprehensive quality assessment.

**Execution + Assessment**: Each trace captures how your application processed a request, while feedback captures how well it performed

**Multi-dimensional Quality**: A single trace can have multiple feedback objects assessing different quality dimensions

**Historical Analysis**: Attached feedback enables tracking quality trends over time and across different application versions

**Debugging Context**: When quality issues arise, you can examine both the execution trace and the assessment rationale

**Answer Completeness**: How thoroughly the response addresses the user's question

**Logical Consistency**: Whether the reasoning and conclusions are sound

Example feedback:
```json
{
  "name": "factual_accuracy",
  "value": 0.85,
  "rationale": "The response correctly identifies 3 out of 4 key facts about MLflow, but incorrectly states the founding year."
}
```

**Context Utilization**: Whether retrieved documents or provided context were used effectively

**Topic Adherence**: Staying on-topic and avoiding irrelevant tangents

Example feedback:
```json
{
  "name": "relevance_to_query",
  "value": "HIGH",
  "rationale": "Response directly answers the user's question about MLflow features and provides relevant examples."
}
```

**Guideline Adherence**: Following specific organizational or ethical guidelines

**Bias Detection**: Identifying unfair bias or discrimination in responses

Example feedback:
```json
{
  "name": "is_safe",
  "value": true,
  "rationale": "Content contains no harmful, toxic, or inappropriate material."
}
```

**Tone Appropriateness**: Whether the tone matches the intended context

**Helpfulness**: How useful the response is to the user

Example feedback:
```json
{
  "name": "helpfulness",
  "value": 4,
  "rationale": "Response provides clear, actionable information but could include more specific examples."
}
```

**Post-Execution Evaluation**: Feedback is generated by evaluating the trace data (inputs, outputs, intermediate steps)

**Multiple Evaluators**: Different evaluation methods can assess the same trace, creating multiple feedback objects

**Batch or Real-time**: Feedback can be generated immediately or in batch processes

**Persistent Storage**: Feedback is stored alongside trace data in MLflow's backend

**Metadata Preservation**: All context about the evaluation method and timing is maintained

**Version Tracking**: Changes to feedback or re-evaluations are tracked over time

**Trend Analysis**: Historical feedback enables tracking quality changes over time

**Comparative Analysis**: Compare feedback across different model versions, prompts, or configurations

**Reporting**: Generate quality reports and dashboards from aggregated feedback data

{
  "name": "comprehensive_quality",
  "value": {
    "overall_score": 0.85,
    "accuracy": 0.9,
    "fluency": 0.8,
    "confidence": 0.75
  }
}

**Advantages**:
- Scale to large volumes of data
- Evaluate subjective criteria
- Provide detailed reasoning
- Consistent evaluation criteria

**Use Cases**:
- Content quality assessment
- Relevance evaluation
- Style and tone analysis
- Complex reasoning evaluation

**Example**: An LLM judge evaluating response helpfulness with detailed rationale explaining specific strengths and weaknesses.

**Advantages**:
- Deterministic and consistent
- Fast execution
- Objective measurement
- Easy to understand and debug

**Use Cases**:
- Format validation
- Length constraints
- Keyword presence/absence
- Quantitative metrics

**Example**: Checking if a response contains required elements or meets length requirements.

**Advantages**:
- Highest quality for subjective criteria
- Nuanced understanding
- Ground truth establishment
- Complex context evaluation

**Use Cases**:
- Quality benchmarking
- Edge case evaluation
- Sensitive content review
- Final quality validation

**Example**: Human reviewers assessing response quality using standardized rubrics and providing detailed feedback.