Feedback Scores¶

Feedback scores allow you to attach quality metrics to traces and spans. Use them to:

• Record user feedback (thumbs up/down, ratings)
• Store evaluation results from automated metrics
• Track quality metrics over time

Adding Feedback to Traces¶

// Add a numeric score
trace.AddFeedbackScore(ctx, "accuracy", 0.95, "High accuracy response")

// Add multiple scores
trace.AddFeedbackScore(ctx, "relevance", 0.87, "Mostly relevant")
trace.AddFeedbackScore(ctx, "helpfulness", 0.92, "Very helpful")

Adding Feedback to Spans¶

// Add feedback to specific spans
span.AddFeedbackScore(ctx, "latency_score", 0.75, "Response time acceptable")
span.AddFeedbackScore(ctx, "quality", 0.90, "Good quality output")

Score Parameters¶

Parameter	Type	Description
name	string	Name of the score (e.g., "accuracy", "relevance")
value	float64	Score value (typically 0.0 to 1.0)
reason	string	Optional explanation for the score

Use Cases¶

User Feedback¶

func handleFeedback(ctx context.Context, traceID string, rating int) {
    // Convert 1-5 rating to 0-1 score
    score := float64(rating-1) / 4.0

    trace := getTrace(traceID)
    trace.AddFeedbackScore(ctx, "user_rating", score,
        fmt.Sprintf("User rated %d/5", rating))
}

Automated Evaluation¶

func evaluateResponse(ctx context.Context, trace *opik.Trace, response string) {
    // Run evaluation metrics
    relevanceScore := evaluateRelevance(response)
    factualScore := evaluateFactuality(response)

    // Record as feedback scores
    trace.AddFeedbackScore(ctx, "relevance", relevanceScore, "Automated relevance check")
    trace.AddFeedbackScore(ctx, "factuality", factualScore, "Automated fact check")
}

A/B Testing¶

func recordABResult(ctx context.Context, trace *opik.Trace, variant string, converted bool) {
    score := 0.0
    if converted {
        score = 1.0
    }

    trace.AddFeedbackScore(ctx, "conversion", score,
        fmt.Sprintf("Variant %s, converted: %v", variant, converted))
}

Viewing Feedback Scores¶

Feedback scores are visible in the Opik UI:

• On the trace detail page
• In trace list summaries
• In experiment comparisons
• In analytics dashboards

Best Practices¶

1. Use consistent names: Standardize score names across your application
2. Normalize values: Use 0.0-1.0 range for easy comparison
3. Include reasons: Add explanations for debugging and analysis
4. Score at appropriate level: Use trace-level for overall quality, span-level for specific operations